ArticlePDF Available

Abstract and Figures

The Corn Belt of the United States, one of the most agriculturally productive regions in the world, experienced a globally anomalous decrease in annual temperatures and a concurrent increase in precipitation during the mid- to late-20th century. Here, we quantify the impact of this ‘warming hole’ on maize yields by developing alternative, no warming hole, climate scenarios that are used to drive both statistical and process-based crop models. We show that the warming hole increased maize yields by 5%–10% per year, with lower temperatures responsible for 62% of the simulated yield increase and greater precipitation responsible for the rest. The observed cooling and wetting associated with the warming hole produced increased yields through two complementary mechanisms: slower crop development which resulted in prolonged time to maturity, and lower drought stress. Our results underscore the relative lack of climate change impacts on central US maize production to date, and the potential compounded challenge that a collapse of the warming hole and climate change would create for farmers across the Corn Belt.
This content is subject to copyright. Terms and conditions apply.
Environ. Res. Lett. 14 (2019)114008 https://doi.org/10.1088/1748-9326/ab422b
LETTER
Mid-20th century warming hole boosts US maize yields
Trevor F Partridge
1,4
, Jonathan M Winter
1,2
, Lin Liu
3
, Anthony D Kendall
3
, Bruno Basso
3
and
David W Hyndman
3
1
Department of Earth Sciences, Dartmouth College, Hanover, NH, United States of America
2
Department of Geography, Dartmouth College, Hanover, NH, United States of America
3
Department of Earth and Environmental Sciences, Michigan State University, East Lansing, MI, United States of America
4
Author to whom any correspondence should be addressed.
E-mail: trevor.f.partridge.gr@dartmouth.edu
Keywords: climate impacts, warming hole, agriculture, machine learning
Supplementary material for this article is available online
Abstract
The Corn Belt of the United States, one of the most agriculturally productive regions in the world,
experienced a globally anomalous decrease in annual temperatures and a concurrent increase in
precipitation during the mid- to late-20th century. Here, we quantify the impact of this warming
holeon maize yields by developing alternative, no warming hole, climate scenarios that are used to
drive both statistical and process-based crop models. We show that the warming hole increased maize
yields by 5%10% per year, with lower temperatures responsible for 62% of the simulated yield
increase and greater precipitation responsible for the rest. The observed cooling and wetting associated
with the warming hole produced increased yields through two complementary mechanisms: slower
crop development which resulted in prolonged time to maturity, and lower drought stress. Our results
underscore the relative lack of climate change impacts on central US maize production to date, and the
potential compounded challenge that a collapse of the warming hole and climate change would create
for farmers across the Corn Belt.
Introduction
The remarkable increase in the United Statesagricul-
tural productivity during the 20th century (Dimitri
et al 2005)occurred largely against the backdrop of a
long-term global climate anomaly. During the second
half of the twentieth century, growing season tempera-
tures throughout most of the central US experienced
an anomalously cool period associated with a
phenomenon commonly referred to as the US warm-
ing hole (Hartmann et al 2013)as well as up to a 35%
increase in summer precipitation (Alter et al 2018).
This cooling and wetting may have created a more
agriculturally favorable climate over the central US as
the 20th century progressed. Given the global impor-
tance of US agriculture, the uncertainty in how future
climate change may impact agricultural productivity,
and the likelihood that the cooling trends associated
with US warming holewill disappear (Kumar et al
2013, Meehl et al 2015), understanding the historic
relationship between yield and climate variability is
critical to current and future adaptation strategies.
Here, we quantify the effects of the US warming hole
on maize, which is one of the most important crops
grown in the central US (i.e. the Corn Belt).
The US warming hole is characterized by an
abrupt shift toward cooler annually averaged tempera-
tures in 1958, followed by a prolonged cool period
over much of the central or southeastern US, depend-
ing on season (Rogers 2013, Mascioli et al 2017,
Partridge et al 2018). The exact timing and spatial
extent of the warming hole varies with season, shifting
from the southeastern US in winter and spring to the
central US in summer and autumn. From 1980 until
present, minimum temperatures throughout the US
have trended upward, however there has been no sig-
nicant trend in maximum temperatures within the
warming hole region (USGCRP 2017). The mechan-
isms responsible for the observed regime shift and
resultant lack of warming trend also vary by season
and region. Studies focusing on the southeastern US
suggest that the observed cooling trend is likely a result
OPEN ACCESS
RECEIVED
23 May 2019
REVISED
23 August 2019
ACCEPTED FOR PUBLICATION
6 September 2019
PUBLISHED
22 October 2019
Original content from this
work may be used under
the terms of the Creative
Commons Attribution 3.0
licence.
Any further distribution of
this work must maintain
attribution to the
author(s)and the title of
the work, journal citation
and DOI.
© 2019 The Author(s). Published by IOP Publishing Ltd
of internal variability in the climate system, such as
large-scale oscillations in the Pacic and Atlantic
oceans; (Meehl et al 2012, Partridge et al 2018),
increasing aerosol emissions during the early and mid-
twentieth century (Leibensperger et al 2012, Mascioli
et al 2017), and regional reforestation (Misra et al
2012, Ellenburg et al 2016). In the central US, agri-
cultural practices have been linked to the observed
summertime cooling trends. Increased irrigation
(Lobell and Bonls 2008, Pei et al 2016, Deines et al
2017)and agricultural intensication (Mueller et al
2015, Alter et al 2018)have both been shown to have
remarkable inuence on regional temperature and
precipitation patterns. In this paper, we focus on the
growing season of maize in the US Corn Belt, and thus
use the term warming hole in reference to the sum-
mertime cooling pattern observed over the central US
(gure 1).
Between 1930 and 2011, the yields within the aver-
age maize producing county in the US increased from
3.02 to 12.43 t ha
1
, with annual growth rates peaking
at 3%5% during the 1960s (Kucharik and Raman-
kutty 2005). Although increased fertilizer applications,
improved cultivar genetics, and innovations in agri-
cultural technology are responsible for much of the
yield improvement, decreased interannual climate
variability (Thompson 1986, Baker et al 1993)and
prolonged growing seasons with reduced heat
extremes since 1980 (Butler et al 2018)have also con-
tributed to the impressive maize yield gains.
Here, we investigate the relationship between maize
yields and climate through a multi-model approach,
combining results from a process-based and statistical
crop model. Process-based crop models can capturethe
physiological response of crops to variations in climate
and management conditions because they explicitly
simulate plant growth and development throughout
the growing season. Process-based models have been
extensively validated globally (Basso et al 2016), how-
ever, they require site-specic input on soil, climate,
genotypes and management which can often be limited,
making them more uncertain (Lobell and Asseng 2017).
Statistical crop models are computationally inexpen-
sive, require little to no eld-based calibration data, and
provide reasonable estimates of yield variability (Lobell
and Burke 2010,Jeonget al 2016, Lobell and Asseng
2017). However, they lack physiological mechanisms
and thus cannot provide the mechanistic insights
offered by process models. We apply two separate mod-
els to simulate long-term rainfed and irrigated maize
yields: the System Approach to Land Use Sustainability
(SALUS)model (Basso et al 2006), and an empirical
machine learning algorithm calibrated to county-level
annual yields. We then statistically remove the warming
hole from the climate record, producing counterfactual
climate scenarios to drive the crop models. Finally,
we analyze the scenario outputs and quantify the net
yield benet provided by the warming hole, as well
as the proportional benets provided individually by
the reduced temperatures and increased precipitation
within the warming hole.
Methods
Aowchart of the methodology described below is
included in the supporting information as gure S1 is
available online at stacks.iop.org/ERL/14/114008/
mmedia.
Data
We acquire historical 19302011 county level yield,
production, planting date, and plant growth stage data
for maize from the US Department of Agriculture
National Agricultural Statistical Service (USDA
NASS). Climate data is derived from daily gridded
Figure 1. Growing season temperature trends and average annual maize production. (a)County level trend in growing season average
temperature from 1930 to 2011 over cultivated land. Counties with no cultivated land are masked in grey. Growing season varies by
state based on USDA NASS planting data (approximately early May to September for Corn Belt).(b)Average annual 1990 to 2011
maize production in metric tons by county. Domain of focus (82°Wto102°Wand 37°Nto47°N)is identied by the black box.
2
Environ. Res. Lett. 14 (2019)114008
values of maximum temperature (Tmax), minimum
temperature (Tmin), daily precipitation (Pr), and
incoming shortwave radiation from the Livneh hydro-
meteorological dataset (Livneh et al 2013,2015). The
Livneh hydrometeorological dataset contains meteor-
ological observations from 20 000 weather stations
across the US and ux estimates gridded to a 1/16°
scale from 1915 to 2015 for the continental US and
parts of Canada and Mexico. Within Livneh, short-
wave radiation is estimated using the MTCLIM algo-
rithm in the Variable Inltration Capacity model
(Maurer et al 2002). We note that our analysis ends in
2011 because Livneh shortwave radiation data are only
available for 19152011. We approximate county
irrigation by the fraction of average historical
19302005 area equipped for irrigation (AEI, Siebert
et al 2014)relative to a countys cultivated area
(cultivated area calculation explained below).
Gridded Tmax, Tmin, and Pr values are aggre-
gated to the county level for the random forest. To
reduce the inuence of non-representative grid cells
on county aggregates, such as cells covering mountai-
nous or metropolitan areas, we overlay the gridded cli-
mate data with a cultivated area mask developed using
the 2017 Cultivated Layer dataset from USDA-NASS
(Boryan et al 2012). The 2017 National Cultivated
Layer uses national cropland data layers from 2013 to
2017 to estimate cultivated area at 30 m binary resolu-
tion across the United States. We interpolate the culti-
vated layer to a 1/16°grid, and aggregate climate data
to the county level using grid cells corresponding with
cultivated land. As total land area devoted to agri-
culture has decreased by more than 400 000 km
2
dur-
ing the 20th century (Brown et al 2005; World
Bank 2019)we argue that using recent cultivated area
estimates are likely conservative.
Removing the warming hole from the climate record
We develop counterfactual (no warming hole)scenar-
ios using two techniques: a delta-change method
(hereafter Delta)and a quantile mapping method
(hereafter QMap). Both methods are commonly used
to bias correct global climate model data for use in
regional assessments (Maraun and Widmann 2018,
Winter et al 2019). The QMap approach produces a
nearly identical shift in the mean value as the Delta
approach, but also adjusts the variance of the post-
1958 distribution to pre-1958 values (gure S2). Thus,
including both methods allows for a direct compar-
ison of how a mean shift in climate affects maize yields
relative to a shift in both the mean and variance.
The Delta approach effectively adjusts observa-
tions by some change factor or delta, calculated as the
difference between a target and reference distribution,
additively for temperature and multiplicatively for
precipitation. We remove the observed 1958 regime
shift associated with the onset of the warming hole
(Rogers 2013, Partridge et al 2018)by adjusting daily
values of Tmax, Tmin, Pr, and SW by the difference or
ratio between the pre-1958 (19151957)and post-
1958 (19592011)long-term monthly means. This
approach preserves interannual variability of the post-
1958 data but adjusts the mean to pre-1958 values
(gure S2).
The QMap scenario is constructed in four steps
(Brocca et al 2011). First, we divide Tmax, Tmin, Pr,
and SW into pre- and post-1958 distributions. Second,
we nd the difference between the pre- and post-1958
values for each post-1958 percentile. Third, we ta
polynomial function to the post-1958 values and the
calculated differences between the pre- and post-1958
values by percentile. Finally, we adjust the post-1958
values using the calculated differences from the poly-
nomial function. For both methods, we mandate that
all adjusted values lie within the bounds of historical
records. Any values (less)greater than the maximum
(minimum)historical value are set to the maximum
(minimum)historical value and any negative values of
Pr and SW are set to 0.
In addition to the Delta and QMap scenarios, we
create three additional scenarios to address the sensi-
tivity of the model outcomes to the time period inclu-
ded. To account for the inclusion of the Dust Bowl, an
extraordinarily warm time period, in the reference dis-
tribution, we create a separate counterfactual climate
scenario with the Dust Bowl years (19301939)
removed from the reference period. Additionally, we
test the sensitivity of onset of the warming hole by
creating two counterfactual scenarios with break-
points at 1953 and 1963, 5 years earlier and later than
the observed warming hole onset, respectively.
Model framework
In this study, we use the well-validated SALUS model
(Basso et al 2006)and the Treebagger random forest
algorithm in Matlab (Breiman 2001)to simulate long-
term rainfed and irrigated maize yield. Both models
are calibrated to county-level reported yields from
19302011 (random forest)and 19562011 (SALUS).
A detailed description of the SALUS model can be
found in the supplemental text S1. SALUS simulations
do not extend earlier than 1956 because this would
require assumptions based on limited to no cultivar
and management data. Our analysis spans the Corn
Belt, from 82°Wto102°W and 37 °Nto47°N. To
ensure that this analysis includes predominantly
agricultural counties, any county with less than 30
years of yield data or fewer than 10 000 harvested acres
on average is excluded, resulting in 794 counties. For
both models, we assume static planting dates within
each state, set to the 19792011 average statewide
median reported planting date. The length of the
growing season is also static for the random forest, as
described below. However, maize harvest dates in the
SALUS model vary dynamically by state and year to
capture the maize growth cycle at maturity.
3
Environ. Res. Lett. 14 (2019)114008
For SALUS validation, we simulate both rainfed
(counties with AEI fraction <5%; n=590)and irri-
gated (counties with AEI fraction >60%; n=4, all in
Nebraska)maize yields. Automatic irrigation is trig-
gered when plant available water content drops below
50% of maximum soil water holding capacity. After
triggering, water is then added to the soil until plant
available water content reaches 70% of maximum soil
water holding capacity. We account for some of the
technological changes in the simulated 56 years by
varying management practices. The crop genetic coef-
cients are obtained using an inverse calibration pro-
cedure to match state-wide average reported yields to
yields from maize cultivar and management practices
used in the 56 year simulated period. Specically, we
vary maize cultivars from low yielding maize cultivars
in 1950s to high yielding cultivars in recent decades,
planting densities from 4 to 10 plants m
2
, and nitro-
gen fertilizer application rates from less than 10 to
120 kg ha
1
. We extract the dominant soil informa-
tion for each map unit polygon in the Soil Survey Geo-
graphic Database as soil input for each studied county
(USDA/NRCS 2018)and ran a unique SALUS simula-
tion for each cell of the 1/16°climate grid. For each
simulation year and grid cell, we calculate the AEI-
weighted average yield of the simulated rainfed and
irrigated yields. We then average the weighted annual
yield across the grid cells in each county.
SALUS allows us to quantify the mechanistic
effects of the warming hole on altered yields. Here we
use both time to maturity, calculated directly by
SALUS, as well as a drought stress metric that is the
average of a daily drought stress factor over the 20-day
period from vegetative-tasseling to early grain lling,
as dened by SALUS. The drought stress factor is
computed as
=- /
F
WW1
,
drought supply demand
where
W
supply is the total water available to the plant and
W
demand
is the potential evaporative demand, calcu-
lated using the Ritchie equations (Ritchie 1972). Thus,
values of
=
F
1
drought
indicates complete drought
stress, and
=
F
0
drought
indicates no drought stress.
Random forest is a non-parametric supervised
ensemble-based machine learning algorithm consist-
ing of a large number of individual decision trees
trained on random subsets of the input data and pre-
dictor variables (Breiman 2001). Random Forest algo-
rithms have been shown to consistently outperform
multiple linear regression methods in applications
including climate and agriculture (Jeong et al 2016).In
this study, we train a random forest of 700 trees on
19302011 county yield data for the 794 counties
using nine predictor variables; three non-climate and
six climate. For each decision tree, we mandate the leaf
size, or the number of observations per leaf (each split
in the decision tree produces two leaves), to be more
than 10 to avoid overtting. The non-climate variables
include: (1)year, (2)average county latitude, and (3)
the average historical fraction of cultivated area that is
equipped for irrigation (AEIFrac). The six climate
variables are: (1)growing season growing degree days
(GDD, see supplemental text S2),(2)growing season
killing degree days (KDD; see supplemental text S2;
Butler and Huybers 2015),(3)total growing season
precipitation (Pr
gs
),(4)average temperature during a
critical period of the growing season (Tavg
Cr
),(5)
maximum Tmax during a critical period of the grow-
ing season (Txx
Cr
), and (6)total Pr during a critical
period of the growing season (Pr
Cr
). These predictor
variables are chosen from a full suite of climate metrics
by highest permuted predictor importance.
For the random forest model, the growing season
is dened as the period from the planting date (descri-
bed above)to the last day of the week corresponding
with at least 50% of a states corn reaching maturity
based on average 19792011 NASS survey data
(USDA/NASS 2018). We identify a county-specic
critical period within the growing season following a
similar approach as Teasdale and Cavigelli (2017).We
correlate 19302011 county yield timeseries with
weekly aggregates of Tavg and Pr for all combinations
of consecutive weeks from the rst week of a countys
growing season to the last week of the growing season,
for each year in the dataset. A county-specic critical
period is dened as the period of weeks that returns
the highest absolute correlation between yield and
either Tavg or Pr. We then dene the critical period for
the model as the average of those from the 794 coun-
ties; here weeks 1217 post planting for Tavg
Cr
and
Txx
Cr
and 912 weeks post planting for Pr. A graphical
representation of the correlation matrix identifying a
countys critical period is shown in gure S3.
To evaluate both modelsaccuracy at simulating
historical maize yields, we use the root mean square
error and the mean absolute percentage error as well as
the coefcient of determination (r
2
), the p-value, the
slope, and the intercept of a linear regression model
between simulated and observed yield. For evaluating
the random forest, we use out-of-bag predictions that
produce unbiased metrics without needing to divide
the observations into training and test sets (Brei-
man 1996). Correlation plots of the observed versus
simulated yields are shown in gure S4.
Results
Both SALUS and the random forest simulate maize
yields that closely reproduce the reported yields (gure
S4). Table 1lists the model performance evaluation
metrics. The high evaluation metrics for the random
forest are due to our decision to train the model on
strongly-trending annual yield data while including
year as a predictor variable. Training the random
forest model with year as the only predictor variable
illustrates the signicance of this decision (gure
S4(c)). However, importantly, aspects of interannual
climate variability (e.g. large drought events)are still
captured by this simplied model, and we argue that
4
Environ. Res. Lett. 14 (2019)114008
this approach is analogous to the common approach
of training a model on detrended yield data (e.g. Lobell
and Field 2007)given that the relationship between
year and yield within the full random forest model is
nearly linear (gure S5). In addition, using yields that
are not detrended allows for the direct comparison of
results between the random forest and SALUS crop
model.
SALUS and the random forest model results agree
that maize yields beneted from the warming hole.
Figure 2illustrates the effect of removing the warming
hole on Tmax and the corresponding impact on pre-
dicted maize yield for Green County, IL. Average
growing season Tmax is approximately 1.5 °C higher
under the no warming-hole scenario in Greene
County and the corresponding predicted yield decrea-
ses by an average of 26%.
Without the warming hole, over fty percent of
counties in the Corn Belt would have experienced
average annual maize yields at least 5.2%9.8% below
historical values, depending on the climate scenario
and model (gure 3). We report the yield difference as
the average annual percent difference between the pre-
dicted yield under the historical climate and a given
counterfactual climate scenario. Thus, a positive yield
difference represents a yield benet from the observed
warming hole. The largest yield benets occur under
the QMap scenario, with median differences of 9.8%
for the random forest and 7.8% for SALUS. Counties
in the Dakotas and western Minnesota exhibit the lar-
gest yield differences of up to 20% for SALUS simula-
tions and 24% according to the random forest,
suggesting that this area may have beneted the most
from the warming hole. The models forced with the
Delta scenario indicate that fty percent of counties
would have experienced at least a 5.2% (random for-
est)and 7.2% (SALUS)reduction in yield without the
warming hole (gure S6). The QMap and Delta sce-
narios have a similar shift in mean climate, so the dif-
ference in median yield response between the two
scenarios is effectively the yield response due to a
change in climate variability alone. Surprisingly, the
random forest is more sensitive than SALUS to the
increased variability within a growing season as
imposed under the QMap scenario even though it uses
primarily on growing season aggregate climate data.
These results are qualitatively robust to the breakpoint
used in creating the counterfactual scenarios and the
exclusion of the Dust Bowl in the reference period.
Perhaps unsurprisingly, removing the Dust Bowl
from the reference period reduces the median yield
benet to 4.6% for the random forest forced by the
QMap scenario, however the vast majority (>96%)of
counties still exhibit increased yields relative to the
Table 1. Model evaluation metrics for SALUS and the random forest models. Metrics include root mean square
error (RMSE)and mean annual percent error (MAPE).
Observed versus predicted linear regression
RMSE (tha
1
)MAPE (%)Intercept (tha
1
)Slope r
2
p-value
SALUS 1.44 19.3 1.22 0.81 0.67 <0.01
Random forest 0.71 15.9 0.18 1.03 0.94 <0.01
Figure 2. Effect of removing the warming hole on Tmax and corresponding impact on yield.(a)Time series of maximum temperature
for Greene County IL, under historical climate and no-warming hole counterfactual scenario developed with quantile mapping.
(b)Time series of reported yield for Greene County, IL and predicted yield using the random forest algorithm under historical climate
and counterfactual no warming hole climate developed with quantile mapping. Predicted yield under historical climate (black line)is
estimated using out-of-bag (OOB)prediction.
5
Environ. Res. Lett. 14 (2019)114008
counterfactual scenario with the Dust Bowl removed
(not shown).
In addition to forcing the models with the full
counterfactual climate scenarios, we also explore the
isolated effects of temperature and precipitation on
maize yields (gure 3(c)). Crop models cannot attri-
bute the simulated yield benet to one particular fac-
tor, given the complex nature of the soil-plant-
atmosphere interactions and feedbacks, so we quantify
the relative importance of temperature and precipita-
tion using the two additional counterfactual scenarios
derived from the QMap scenario. QMap
T
and QMap
Pr
represent historical observations of shortwave radia-
tion with either Tmin and Tmax or precipitation
adjusted by quantile mapping, respectively. The ran-
dom forest suggests that the yield difference under the
QMap scenario is driven almost entirely by the temp-
erature difference between the historical climate and
the QMap climate. For the random forest, there was
no statistical difference in maize yield when precipita-
tion alone was adjusted under the QMap scenario
(p=0.64). These results suggest that the maize yields
in the Corn Belt have beneted little from the increas-
ing summer precipitation in the region, with the possi-
ble exception of the more arid and non-irrigated
northwest corner of the domain (gures S7 and S8).
However, statistical models struggle to capture to
importance of precipitation on plant growth and con-
sequently yield (Lobell and Burke 2010, Ortiz-Bobea
et al 2019).
By contrast, SALUS results suggest that potential
yield gains as a result of the warming hole come from
both lower temperatures and increased precipitation.
SALUS results indicate that 62% of the median yield
benet resulted from the lower temperatures asso-
ciated with the warming hole, while the remaining
benet was a result from increased precipitation
(38%). We identify two potential causes for the differ-
ence in the importance of precipitation between the
random forest and SALUS. First, the random forest
was trained using yield observations from the whole
corn belt, and therefore has a limited ability to simu-
late geographical variations in the sensitivity of maize
to precipitation. Second SALUS is able to simulate the
impacts of daily precipitation on crop growth, while
the random forest aggregates precipitation.
To explore potential mechanisms responsible for the
yield reductions under the no-warming hole climate sce-
narios, we examine differences in simulated drought
stress and maturity time using the SALUS model. Under
Figure 3. Difference in historical maize yield between three climate scenarios and two modeling approaches. (a)Difference in
predicted maize yield from the SALUS crop model between the historical climate and alternative climate scenario with no warming
developed using a quantile mapping method (QMap).(b)Same but using the random forest statistical crop model. (c)Violin plots of
the yield difference for al counties under each climate scenario. Each climate scenario removes the warming hole using a different
statistical method. QMap: temperature and precipitation adjusted by quantile mapping approach; Delta: temperature and
precipitation adjusted by a Delta-change approach; QMap
T
: Only temperature adjusted by quantile mapping approach; QMap
Pr
:
Only precipitation adjusted by quantile mapping approach.
6
Environ. Res. Lett. 14 (2019)114008
both no warming hole scenarios, drought stress increases,
and crops mature earlier, leading to lower yields. Maize in
the average county reaches maturity 2.7 d or 2.9 d earlier
than the historical observed climate and experiences an
52% or 49% increase in drought stress for the QMap or
the Delta scenarios, respectively (gure 4).Thelargestdif-
ference in crop maturity time occurs in the northern
counties of Wisconsin where cooler late spring and early
autumn temperatures otherwise slow plant development.
Interestingly, the northern counties of Minnesota and the
Dakotas do not exhibit the same reduction in time to
maturity, likely due to later planting dates in this region.
There is a longitudinal gradient in the reduction of
drought stress between the historical and QMap scenar-
ios. Counties in the eastern half of the domain appear to
have substantially lower drought stress levels as a result of
the warming hole and concurrent precipitation trends.
Random forest algorithms are often referred to as
black boxes, as it can be difcult to interpret the rela-
tionships formed within the ensemble of decision
trees. Accumulated local effect (ALE)plots provide a
way to visualize the relationship between the response
variable (yield)and each predictor variable in isolation
(Apley 2016). Figure 5shows ALE plots for climate
variables used in the random forest as well as the total
distribution of that variable within the domain. The
plots are centered at zero, which represents the mean
model prediction across all values of the variable.
Values below zero have a net negative effect on yield
(decreased simulated yield)and values above zero are
benecial (increased simulated yield). ALE plots for
the non-climatic variables are shown in gure S5.
GDD, the hottest daily maximum temperature during
the critical period (Txx
cr
),1217 weeks after planting,
and KDD (Butler and Huybers 2015)have a similar
effect on yield (see methods for description of random
forest and variable calculation). The ALE plot of aver-
age temperature during the critical period (Tavg
cr
)
suggests that days with average temperature above
22 °C can be detrimental to maize yields. However,
this threshold should be interpreted with caution as
Tavg is a statistical construct of daily Tmax and Tmin,
which are more likely to physically affect plant growth.
The ALE plots of Pr give insight into why forcing the
random forest with QMap
pr
had little effect on simu-
lated yield (gure 3(c)). Precipitation summed over
the growing season (Pr
gs
)and during the critical per-
iod (Pr
cr
; weeks 912 post planting)exhibit relatively
at curves, and the corresponding precipitation dis-
tributions illustrate the general abundance of water in
the region. The more arid northwestern portion of the
domain did benet from the historical increase in
summer precipitation, so a similar analysis focused
only on this region might nd a stronger relationship
between Pr and maize yields (gure S7).
Conclusion
Understanding the dynamic relationship between
agricultural productivity and regional climate is of
critical importance given the challenges imposed by
population growth, shifting diets, and climate change.
Here we nd that the 20th century trend in US maize
yield was signicantly aided by anomalously cool
temperatures and increased growing season precipita-
tion. These results expand upon those of Butler et al
(2018), who use a statistical crop model to show that
increased minimum temperatures and a reduction in
maximum temperature extremes since 1980 have been
benecial to maize yields. We nd that this agricultu-
rally pleasant climate extends back to 1958 and also
quantify the importance of a simultaneous increase in
growing season precipitation. We further identify both
prolonged maturity time and reduced water stress as
the physiological mechanisms responsible for the yield
benet(gure 4). It is unlikely that the pleasant climate
Figure 4. Impacts of warming hole on time to maturity and drought stress. (a)Difference in time to maturity of SALUS simulated
maize yield between historical and QMap climate scenario. (b)Difference in drought stress between historical climate and QMap
scenario. Positive values of time to maturity indicate later maturity, and negative drought stress values indicate reduced drought stress,
in the presence of the warming hole.
7
Environ. Res. Lett. 14 (2019)114008
observed during the latter half of the 20th century will
substantially persist into the future. Although max-
imum growing season temperatures in the Corn Belt
continue to increase at a slower rate than minimum
temperatures, possibly due to agricultural expansion
(Mueller et al 2015, Alter et al 2018, Nikiel and
Eltahir 2019), annual temperatures are projected to
increase by 2.3 °C by mid-century (RCP 4.5; Vose et al
2017). Summer precipitation in the region is projected
to decrease by approximately 10% while extreme
events will likely increase (Easterling et al 2017).
The dramatic improvements in technology and
management practices across US Corn Belt produced
remarkable increases in maize yield and productivity.
However, we show that a contemporaneous climate
anomaly referred to as the US warming hole, which
translated to more advantageous weather in the
region, also contributed to yield increases. We nd
that the warming hole resulted in an increased median
annual maize yield of 5%10%, primarily in response
to decreased growing season temperatures and
decreased heat extremes. SALUS driven with counter-
factual scenarios isolating the increase in precipitation
and decrease in temperature indicate that cooler
temperatures were responsible for 62% (and thus
increased precipitation was repsonsible for 38%)of
the simulated yield increase across most of the Corn
Belt. As the central US prepares to adapt to substantial
projected increases in temperatures by the end of the
21st century, it is essential to understand the extent
that the US warming hole has historically benetted
maize yields. Our results underscore the relative lack
of climate change impacts on central US maize pro-
duction to date, and the potential compounded chal-
lenge that a collapse of the warming hole and climate
change would create for farmers across the Corn Belt.
Acknowledgments
This study was funded by the United States Depart-
ment of Agriculture National Institute of Food and
Agriculture (grants 2015-68007-23133 and 2018-
67003-27406), National Science Foundation (BCS
184018), and Nelson A. Rockefeller Center at Dart-
mouth College. This paper makes use of agronomic
data provided by the United States Department of
Agriculture National Agricultural Statistics Service
and the Livneh hydrometeorological climate dataset
provided by NOAA/OAR/ESRL PSD, Boulder, Col-
orado, USA, from their Web site at https://esrl.noaa.
gov/psd/.
Figure 5. Accumulated local effects (ALE)plots of climate variables from the random forest model. Cumulative density functions
show the distribution of each variable. Positive values of ALE indicate a yield benet relative to the mean. Note the distinct vertical axes
scales between the bottom and top rows. ALE plots and histograms are truncated at the 1% and 99% quantiles to remove outliers.
8
Environ. Res. Lett. 14 (2019)114008
Author contributions
TFP, JMW, ADK and DWH designed the study and
developed the random forest model. TFP and JMW
developed the climate scenarios. LL and BB ran the
SALUS model simulations and analyzed drivers of
yield changes. All authors contributed to writing the
manuscript.
Competing interests statement
We have no conicts of interest to declare and none of
the material has been published or is under considera-
tion elsewhere.
ORCID iDs
Trevor F Partridge https://orcid.org/0000-0003-
1589-4783
Jonathan M Winter https://orcid.org/0000-0003-
1261-4774
Lin Liu https://orcid.org/0000-0001-9078-9516
Anthony D Kendall https://orcid.org/0000-0003-
3914-9964
Bruno Basso https://orcid.org/0000-0003-
2090-4616
David W Hyndman https://orcid.org/0000-0001-
9078-9516
References
Alter R E, Douglas H C, Winter J M and Eltahir E A B 2018
Twentieth century regional climate change during the
summer in the central united states attributed to agricultural
intensication Geophys. Res. Lett. 45 158694
Apley D W 2016 Visualizing the effects of predictor variables in
black box supervised learning models arXiv:1612.08468
Baker D G, Ruschy D L and Skaggs R H 1993 Agriculture and the
recent Benign Climatein Minnesota Bull. Am. Meteorol. Soc.
74 103540
Basso B, Dumont B, Cammarano D, Pezzuolo A, Marinello F and
Sartori L 2016 Environmental and economic benets of
variable rate nitrogen fertilization in a nitrate vulnerable zone
Sci. Total Environ. 545546 22735
Basso B, Ritchie J T, Grace P R and Sartori L 2006 Simulation of
tillage systems impact on soil biophysical properties using the
SALUS model Ital. J. Agronomy 1677
Boryan C, Yang Z and Di L 2012 Deriving 2011 cultivated land cover
data sets using usda national agricultural statistics service
historic cropland data layers 2012 IEEE Int. Geoscience and
Remote Sensing Symp. (Piscataway, NJ: IEEE)pp 6297300
Breiman L 1996 Out-Of-Bag Estimation Technical Report (Berkeley,
CA: Department of Statistics, University of California,
Berkeley)(https://www.stat.berkeley.edu/~breiman/
OOBestimation.pdf)
Breiman L 2001 Random forests Mach. Learn. 45 532
Brocca L, Hasenauer S, Lacava T, Melone F, Moramarco T,
Wagner W and Bittelli M 2011 Soil moisture estimation
through ASCAT and AMSR-E sensors: an intercomparison
and validation study across Europe Remote Sens. Environ. 115
3390408
Brown D G, Johnson K M, Loveland T R and Theobald D M 2005
Rural land-use trends in the conterminours United Sates,
19502000 Ecol. Appl. 15 185163
Butler E E and Huybers P 2015 Variations in the sensitivity of US
maize yield to extreme temperatures by region and growth
phase Environ. Res. Lett. 10 034009
Butler E E, Mueller N D and Huybers P 2018 Peculiarly pleasant
weather for US maize PNAS 115 1193540
Deines J, Kendall A and Hyndman D W 2017 Annual irrigation
dynamics in the US Northern High Plains derived from
Landsat satellite data Geo. Res. Lett. 44 935060
Dimitri C, Efand A and Conklin N 2005 The 20th Century
Transformation of US Agriculture and Farm Policy (Economic
Information Bulletin No. 3)(Washington, DC: USDA Eco-
nomic Research Service)(www.ers.usda.gov)
Easterling D R, Arnold J R, Knut son T, Kunkel K E, LeGra nde A N,
Leung L R and Wehner M F 2017 Precipit ation Change in the
United States. Climate Science Special Report: Fourth
National Climate Assessment ed D J Wueebbles et al vol 1
(Washington , DC: US Global Change R esearch Program)
pp 20730
Ellenburg W L, McNider R T, Cruise J F and Christy J R 2016
Towards an understanding of the twentieth-century cooling
trend in the Southeastern United States: biogeophysical
impacts of land-use change Earth Interact. 20 131
Hartmann D J, Klein Tank A M G, Rusticucci M, Alexander L V,
Brönnimann S, Charabi Y A R and Zhai P 2013 Observations:
Atmosphere and Surface Climate Change 2013: The Physical
Science Basis. Contribution of Working Group I to the Fifth
Assessment Report of the Intergovernmental Panel on Climate
Change pp 159254
Jeong J H, Resop J P, Mueller N D, Fleisher D H, Yun K,
Butler E E and Kim S-H 2016 Random forests for global and
regional crop yield predictions PLoS One 11 e0156571
Kucharik C J and Ramankutty N 2005 Trends and variability in
US corn yields over the twentieth century Earth Interact. 9
129
Kumar S, Merwade V, Kinter J L and Niyogi D 2013 Evaluation of
temperature and precipitation trends and long-term
persistence in CMIP5 twentieth-century climate simulations
J. Clim. 26 416885
Leibensperger E M, Mickley L J, Jacob D J, Chen W T, Seinfeld J H,
Nenes A and Rind D 2012 Climatic effects of 19502050
changes in US anthropogenic aerosols: II. Climate response
Atmos. Chem. Phys. 12 334962
Livneh B, Bohn T J, Pierce D W, Munoz-Arriola F, Nijssen B,
Vose R and Brekke L 2015 A spatially comprehensive,
hydrometeorological data set for Mexico, the US, and
Southern Canada 19502013 Sci. Data 2150042
Livneh B, Rosenberg E A, Lin C, Nijssen B, Mishra V,
Andreadis K M and Lettenmaier D P 2013 A long-term
hydrologically based dataset of land surface uxes and states
for the conterminous United States: update and extensions
J. Clim. 26 938492
Lobell D B and Asseng S 2017 Comparing estimates of climate
change impacts from process-based and statistical crop
models Environ. Res. Lett. 12 015001
Lobell D B and Bonls C 2008 The effect of irrigation on regional
temperatures: a spatial and temporal analysis of trends in
California, 19342002 J. Clim. 21 206371
Lobell D B and Burke M B 2010 On the use of statistical models to
predict crop yield responses to climate change Agric. For.
Meteorol. 150 144352
Lobell D B and Field C B 2007 Global scale climatecrop yield
relationships and the impacts of recent warming Environ. Res.
Lett. 2014002
Maraun D and Widmann M 2018 Statistical Downscaling and Bias
Correction for Climate Research (Cambridge: Cambridge
University Press)
Mascioli N R, Previdi M, Fiore A M and Ting M 2017 Timing and
seasonality of the United States warming holeEnviron. Res.
Lett. 12 034008
Maurer E P, Wood A W, Adam J C, Lettenmaier D P and Nijssen B
2002 A long-term hydrologically based dataset of land surface
uxes and states for the conterminous United States J. Clim.
15 323751
9
Environ. Res. Lett. 14 (2019)114008
Meehl G A, Arblaster J M and Branstator G 2012 Mechanisms
contributing to the warming hole and the consequent US
east-west differential of heat extremes J. Clim. 25
6394408
Meehl G A, Arblaster J M and Chung C T Y 2015 Disappearance of
the southeast US warming holewith the late 1990s transition
of the Interdecadal Pacic Oscillation Geophys. Res. Lett. 42
507481
Misra V, Michael J P, Boyles R, Chassignet E P, Grifn M and
Obrien J J 2012 Reconciling the spatial distribution of the
surface temperature trends in the Southeastern United States
J. Clim. 25 36108
Mueller N D, Butler E E, McKinnon K A, Rhines A, Tingley M,
Holbrook N M and Huybers P 2015 Cooling of US Midwest
summer temperature extremes from cropland intensication
Nat. Clim. Change 631722
Nikiel C A and Eltahir E A B 2019 Summer climate change in the
midwest and great plains due to agricultural development
during the twentieth century J. Clim. 32 558399
Ortiz-Bobea A, Wang H, Carrillo C M and Ault T R 2019 Unpacking
the climatic drivers of US agricultural yields Environ. Res. Lett.
14 064003
Partridge T F, Winter J M, Osterberg E C, Hyndman D W,
Kendall A D and Magilligan F J 2018 Spatially distinct
seasonal patterns and forcings of the US Warming Hole
Geophys. Res. Lett. 45 205563
Pei L, Moore N, Zhong S, Kendall A D, Gao Z and Hyndman D W
2016 Effects of Irrigation on summer precipitation over the
United States J. Clim. 29 354158
Ritchie J T 1972 Model for predicting evaporation from a row crop
with incomplete cover Water Resour. Res. 8120413
Rogers J C 2013 The 20th century cooling trend over the
southeastern United States Clim. Dyn. 40 34152
Siebert S, Kummu M, Porkka M, Döll P, Ramankutty N and
Scanlon B R 2014 A global dataset of the extent of irrigated
land from 1900 to 2005 Hydrol. Earth Syst. Sci. Discuss. 11
1320758
Teasdale J R and Cavigelli M A 2017 Meteorological uctuations
dene long-term crop yield patterns in conventional and
organic production systems Sci. Rep. 7688
Thompson L M 1986 Climatic change, weather variability, and corn
production1 Agron. J. 78 649
USDA NASS 2018 USDANational Agricultural Statistics Service
Quick Stats [Data le](https://nass.usda.gov/Quick_
Stats/)(Retrieved: 26 March 2019)
USDA NRCS 2017 Soil Survey Geographic Database (SSURGO).
NRCS [Data le].(Retrieved: 26 March 2018)
USGCRP 2017 Climate Science Special Report: Fourth National
Climate Assessment ed D J Wuebbles et al vol 1 (Washington,
DC: US Global Change Reseacrh Program)(https://doi.org/
10.7930/J0J964J6)
Vose R S, Easterling D R, Kunkel K E, LeGrande A N and
Wehner M F 2017 Temperature changes in the United States
ed D J Wuebbles et al Climate Science Special Report: Fourth
National Climate Assessment vol 1 (Washington, DC: US
Global Change Research Program)pp 185206
Winter J M, Bowen F L, Partridge T F and Chipman J W 2019 Future
extreme event risk in the rural Northeastern United States
Ann. Am. Assoc. Geogr. 109 111030
World Bank 2019 Agricultural Land (sq. km).[Data le](https://
data.worldbank.org/indicator/AG.LND.AGRI.K2?
locations=US)(
Retrieved: 22 March 2019)
10
Environ. Res. Lett. 14 (2019)114008
... Despite global warming, damaging heat has not increased over the northern Midwest, home to the most productive US maize and soy croplands [6,[22][23][24]. KDD trends inversely covary with yield trends over maize-and soy-growing regions in the US, with areas of strong yield growth having less warming (or cooling) of hot extremes [22,23] (cf figures 1(a)-(f); SI figure 15). ...
... KDD trends inversely covary with yield trends over maize-and soy-growing regions in the US, with areas of strong yield growth having less warming (or cooling) of hot extremes [22,23] (cf figures 1(a)-(f); SI figure 15). Reduced warming trends over croplands have been shown to benefit yields and to be most pronounced on the hottest days of the growing season [6,24,25]. At the same time, GDDs have increased across the entire US (figures 1(c) and (d) and SI figure 2), lengthening the growing season, benefiting maize and soy. ...
Article
Full-text available
U.S. maize and soy production have increased rapidly since the mid-20th century. While global warming has raised temperatures in most regions over this time period, trends in extreme heat have been smaller over U.S. croplands, reducing crop-damaging high temperatures and benefiting maize and soy yields. Here we show that agricultural intensification has created a crop-climate feedback in which increased crop production cools local climate, further raising crop yields. We find that maize and soy production trends have driven cooling effects approximately as large as greenhouse gas induced warming trends in extreme heat over the central U.S. and substantially reduce them over the southern U.S., benefiting crops in all regions. This reduced warming has boosted maize and soy yields by 3.3 (2.7 – 3.9; 13.7 – 20.0%) and 0.6 (0.4 – 0.7; 7.5 – 13.7%) bu/ac/decade, respectively, between 1981 and 2019. Our results suggest that if maize and soy production growth were to stagnate, the ability of the crop-climate feedback to mask warming would fade, exposing U.S. crops to more harmful heat extremes.
... These numerous investigations have relied on empirical/statistical and process-based models, both of which are effective, and should be combined for accurate impact assessments (Ciscar et al. 2018). The findings of these investigations usually report county-specific impacts of temperature (Butler and Huybers, 2015;Shirley et al. 2020;Schauberger et al. 2017;Zhu et al. 2019;Xu et al. 2016;Partridge et al. 2019;Kukal and Irmak, 2018a;Carter et al. 2018;Ortiz-Bobea et al., 2018;Ortiz-Bobea et al., 2019), heat accumulation (Li and Troy, 2018;Butler et al. 2018;Zhu et al. 2018;Kukal and Irmak, 2018b), precipitation (Li and Troy, 2018;Shirley et al. 2020;Zhu et al. 2019;Xu et al. 2016;Kukal and Irmak, 2018a;Carter et al. 2018;Lesk et al. 2020), reference evapotranspiration (Li and Troy, 2018), vapor pressure deficit (Kimm et al. 2020;Hsiao et al. 2019;Lobell et al. 2020;Rigden et al. 2020), moisture availability (Carter et al. 2016;Kimm et al. 2020;Lobell et al. 2020;Rigden et al. 2020;Ortiz-Bobea et al., 2019), solar radiation (Tollenaar et al. 2017;Carter et al. 2018), frost days (Zhu et al. 2019;Kukal and Irmak, 2018b), diurnal temperature range (Zhu et al. 2019;), drought indices (Zipper et al. 2016), elevated CO 2 Urban et al., 2015) and climate extremes (Troy et al. 2015) on maize yield. ...
... e Little is known about how crop phenology (growth phase) dictates the impacts of weather on crop yields. To answer this, static or calendar month-based temporal windows are commonly assumed over large terrestrial regions (Carter et al. 2018;Eggen et al. 2019;Hatfield et al., 2018;Partridge et al. 2019;Ortiz-Bobea et al. 2019). Drastic variability in crop growth and development can exist among individual fields owing to differences in crop establishment dates (a function of spring precipitation and temperature, soil moisture, soil temperature, and days suitable for fieldwork) and latitudinal variation in growing degree days, making it challenging to track crop phenology at large-scales (Butler et al. 2018;Butler and Huybers, 2015;Hoffman et al. 2020;Zhu et al., 2018). ...
Article
Weather impacts on crop productivity have emerged as a significant research area, and agroecosystems globally have been evaluated for these impacts. However, only a limited body of research relies on experimental determination of field-scale impacts, relative to coarser-scale estimations. Experimental frameworks can control confounding factors that potentially undesirably affect coarser-scale assessments. Here, we employ a long-term (31-year) experimental trial to investigate weather-related signatures on rainfed maize (Zea Mays L.) yield in the western Corn Belt, at seasonal and sub-seasonal scales. Among all the weather indices evaluated [(maximum, minimum, and average air temperatures (Tmax, Tmin, Tavg), precipitation (Pcp), incoming shortwave radiation (Rs), wind speed (u2), relative humidity (RH), vapor pressure deficit (VPD), alfalfa-reference evapotranspiration (ETr), and aridity index (AI)], ETr accounted for the most (⁓50%) yield variance, followed by Tavg, Tmax, VPD, Rs, Tmin, Pcp, u2, RH, and AI. Weather indices most effectively accounted for yield variance during the late reproductive (R) phase, followed by early R phase, late vegetative (V) phase and lastly, early V phase. While seasonal ETr was responsible for highest yield sensitivity (-0.69 s.d. s.d.⁻¹), sub-seasonal weather indices were also identified for highest yield sensitivity (u2, Tavg, ETr, Tmax during early V, late V, early R, and late R, respectively). Among all possible combinations, a multiple linear regression model consisting of ETr and Tmin best predicted yield deviations (R²=0.64) at the seasonal scale, while a sub seasonal model consisting of early V ETr, early R ETr, and late R Tmin performed the best (R²=0.66). Both models were validated for five independent counties to predict USDA estimates of rainfed maize yield deviations. Recent advances in mechanistic understanding of nighttime warming (increasing Tmin) and evaporative demand impacts on crop performance concur with our experimental inference of importance of ETr and Tmin for rainfed maize yields.
... The 2021 Global Report on the Food Crisis, released by the United Nations World Food Programme (WFP), documented that sudden food insecurity soared to the highest known level during the period (WFP, 2021. With the intensification of the food crisis and the ever-increasing population, the response of the agricultural system to climate change has attracted extensive attention (Lu et al., 2015;Partridge et al., 2019). Recent research has indicated that climate variations contributed to approximately-one-third of the global crop yield variability (Ray et al., 2015). ...
Article
Northeast China is an important commodity grain base in China and plays an important role in maintaining national and local food security. The intensifying climate variability and strengthening of extreme weather events threaten crop growth and yields in this region. The dynamic crop model coupled into the land surface models showed advances in simulating crop phenology and crop growth response to climate change, however, which was mainly validated for the United States. By comparing with the satellite observed crop growth cycles, we firstly calibrated the parameters for the Weather Research and Forecasting (WRF) coupled Noah-MP-Crop model. Then, we evaluated the performance of the calibrated WRF-Noah-MP-Crop model with the default crop models and the dynamic vegetation models coupled with WRF by comparing our simulations with corresponding satellite measurements and meteorological observations. Our results indicated the augmentation of the Noah-MP-Crop model in WRF significantly improved the crop phenology simulation at the beginning and ending time of the growing season compared with the dynamic vegetation model using unmanaged crops. The satellite-calibrated crop parameters substantially improved the simulation of crop growth, plant physiology, and biomass accumulation for both corn and soybean. Coupling the localized dynamic crop model into the WRF led to considerable decreases in the simulated mean-absolute-errors (MAEs) and biases of the leaf area index, evapotranspiration, and gross primary production compared with the MODIS observed values. Compared with the statistical yield from each province, the modified crop model underestimated the corn yield from 11.1% to 48.6%, whereas overestimated the soybean yield from 16.5% to 162.6%. The coupled WRF-Noah-MP-Crop model with flexible resolutions holds unmatched advantages in estimating and projecting crop growth and yield under future climate change scenarios, which would be critical in ensuring sustainable agricultural development and maintaining future food security.
... On the use of machine-learning to model soybean yield. In recent years, machine-learning techniques have been successfully applied to predict yields of a variety of crops in different world regions [46][47][48][49] . Consistent with those previous studies, the results presented here demonstrate the good predictive ability of the RF algorithm when applied to soybean, with a RMSEP of 0.35 t ha −1 . ...
Article
Full-text available
The rapid expansion of soybean-growing areas across Europe raises questions about the suitability of agroclimatic conditions for soybean production. Here, using data-driven relationships between climate and soybean yield derived from machine-learning, we made yield projections under current and future climate with moderate (Representative Concentration Pathway (RCP) 4.5) to intense (RCP 8.5) warming, up to the 2050s and 2090s time horizons. The selected model showed high R2 (>0.9) and low root-mean-squared error (0.35 t ha−1) between observed and predicted yields based on cross-validation. Our results suggest that a self-sufficiency level of 50% (100%) would be achievable in Europe under historical and future climate if 4–5% (9–11%) of the current European cropland were dedicated to soybean production. The findings could help farmers, extension services, policymakers and agribusiness to reorganize the production area distribution. The environmental benefits and side effects, and the impacts of soybean expansion on land-use change, would need further research.
... ALE technique has been used in other domains to interpret the effect of predictors variables in the "blackbox" supervised model [32][33][34][35] . From the ALE plots, we found that the threshold values of Ut RI, Ut PI, and AC to differentiate FGR from healthy fetuses are comparable with the clinical guidelines. ...
Article
Full-text available
The objective of the study is to investigate the effect of Nuchal Fold (NF) in predicting Fetal Growth Restriction (FGR) using machine learning (ML), to explain the model's results using model-agnostic interpretable techniques, and to compare the results with clinical guidelines. This study used second-trimester ultrasound biometry and Doppler velocimetry were used to construct six FGR (birthweight < 3rd centile) ML models. Interpretability analysis was conducted using Accumulated Local Effects (ALE) and Shapley Additive Explanations (SHAP). The results were compared with clinical guidelines based on the most optimal model. Support Vector Machine (SVM) exhibited the most consistent performance in FGR prediction. SHAP showed that the top contributors to identify FGR were Abdominal Circumference (AC), NF, Uterine RI (Ut RI), and Uterine PI (Ut PI). ALE showed that the cutoff values of Ut RI, Ut PI, and AC in differentiating FGR from normal were comparable with clinical guidelines (Errors between model and clinical; Ut RI: 15%, Ut PI: 8%, and AC: 11%). The cutoff value for NF to differentiate between healthy and FGR is 5.4 mm, where low NF may indicate FGR. The SVM model is the most stable in FGR prediction. ALE can be a potential tool to identify a cutoff value for novel parameters to differentiate between healthy and FGR.
... Existing observational-and modeling-studies indicated a nonlinear yield response to temperature where yield declines when ambient temperature deviates from an optimal threshold for crop growth (Bassu et al 2014, Ray et al 2015, Lobell and Asseng 2017, Li et al 2019, Partridge et al 2019, Wang et al 2020. Such a nonlinear temperature response, inherited from plants' physiological processes, is critical for process-based crop models (Luo 2011). ...
Article
Full-text available
How maize yield responds to precipitation variability in space and time over broader scales is largely unknown compared with the well-understood temperature response, even though precipitation change is more erratic with greater spatial heterogeneity. Here, we develop a method to quantify the spatially explicit precipitation response of maize yield using statistical data and crop models in the contiguous United States. We find the precipitation responses are highly heterogeneous with inverted-U (40.3%) being the leading response type, followed by unresponsive (30.39 %), and linear increase (28.6%). The optimal precipitation threshold derived from inverted-U response exhibits considerable spatial variations, which is higher under wetter, hotter, and well-drainage conditions but lower under drier and poor-drainage conditions. Irrigation alters precipitation response by making yield either unresponsive to precipitation or having lower optimal thresholds than rainfed conditions. We further find that the observed precipitation responses of maize yield are misrepresented in crop models, with a too high percentage of increase type (59.0% versus 29.6%) and an overestimation in optimal precipitation threshold by ~90 mm. These two factors explain about 30% and 85% of the inter-model yield overestimation biases under extreme rainfall conditions. Our study highlights the large spatial heterogeneity and the key role of human management in the precipitation responses of maize yield, which need to be better characterized in crop modeling and food security assessment under climate change.
... Indeed, a previous study estimated that 28% of the corn yield gain in the Midwest since 1981 is attributable to better weather, of which summer T x cooling accounted for almost half of this beneficial trend 7 . Another study 37 that examined the counterfactual, estimated that warming-hole climate boosted corn yields by~5-10%, mainly due to prolonged maturity time and reduction of water stress 31 . ...
Article
Full-text available
Warming generally leads to increased evaporative demand, altering the amount of water needed for growing crops. For the Midwest, some studies have suggested that reaching yield targets by 2050 will not be possible without additional precipitation or large expansion of irrigation. Here, we show that this claim is not supported by the historical summer climate trends, which indicate that the warming of daily average temperatures is largely driven by increases in minimum temperatures, while maximum temperatures have decreased. This has translated into a net decrease in vapor pressure deficit (VPD) and potential evapotranspiration (PET). With the increasing rainfall, this suggests that crop water deficits have likely become less frequent in the region despite the warming climate. By projecting these trends into 2050 and ancillary use of a crop model, we estimate minor changes in PET that would have minimal effects on corn yields (<6%) under persistence of these trends.
... As a result, maize yields have improved across environments from wellwatered to those with nitrogen and water deficits (Cooper et al., 2014(Cooper et al., , 2020Mueller et al. 2019). The yield potential realized from these genetic and management advancements have been enhanced by 5%-10% owing to favorable changes in climate (Partridge et al., 2019). ...
Article
Full-text available
Increasing temperatures in the U.S. Midwest are projected to reduce maize yields because warmer temperatures hasten reproductive development and, as a result, shorten the grain fill period. However, there is widespread expectation that farmers will mitigate projected yield losses by planting longer season hybrids that lengthen the grain fill period. Here, we ask: a) how current hybrid maturity length relates to thermal availability of the local climate, and b) if farmers are shifting to longer season hybrids in response to a warming climate. To address these questions, we used county‐level Pioneer brand hybrid sales (Corteva Agriscience) across 17 years and 650 counties in 10 Midwest states (IA, IL, IN, MI, MN, MO, ND, OH, SD, WI). Northern counties were shown to select hybrid maturities with growing degree day (GDD˚C) requirements more closely related to the environmentally available GDD compared to central and southern counties. This measure, termed “thermal overlap”, ranged from complete (106%) in northern counties to a mere 63% in southern counties. The relationship between thermal overlap and latitude was fit using split‐line regression and a break point of 42.8° N was identified. Over the 17‐year, hybrid maturities shortened across the majority of the Midwest with only a minority of counties lengthening in select northern and southern areas. The annual change in maturity ranged from ‐5.4 to 4.1 GDD year‐1 with a median of ‐0.9 GDD year‐1. The shortening of hybrid maturity contrasts with widespread expectations of hybrid maturity aligning with magnitude of warming. Factors other than thermal availability appear to more strongly impact farmer decision‐making such as the benefit of shorter maturity hybrids on grain drying costs, direct delivery to ethanol biorefineries, field operability, labor constraints, and crop genetics availability. Prediction of hybrid choice under future climate scenarios must include climatic factors, physiological‐genetic attributes, socio‐economic, and operational constraints.
Article
Observations during 1990–2014 depict concurrent cooling trends in land surface air temperature (LSAT) over central-eastern Eurasia (CEE) and northwestern North America (NWNA), while significant warming trends are observed in other areas across Northern Hemisphere (NH). This lack of surface warming of the continental CEE and NWNA is termed continental warming holes (CWH) here. The ability of the 6th phase of the Coupled Model Intercomparison Project (CMIP6) to reproduce the LSAT trends during 1990–2014 over the continental NH is evaluated, and the evaluation includes all CMIP6 historical simulations from 32 unique models and their multi-model ensemble mean (MMEM). The results show that the performances of the different models vary greatly. GISS-E2–1-G-CC model performs best among the 32 models, considering reproducing the correct sign of the LSAT trends as well as reproducing the trend magnitude. However, the MMEM simulates uniformly significant warming trends in NH and fails to capture the continental warming holes, indicating a role of internal variability in the CWH. The mechanisms of the continental warming holes are further investigated using reanalysis datasets, CMIP6 models and AGCM simulations. GISS-E2–1-G-CC also shows a better simulation ability, including the trends in NH sea level pressure and tropical sea surface temperature than other models. Both reanalysis data and GISS-E2–1-G-CC model show high-pressure trends over mid-high latitudes, which corresponded to enhanced Siberia High and weakened Aleutian Low. These atmospheric circulation changes result in cold advection over the CEE and NWNA regions, offsetting anthropogenic warming and leading to the emergence of the warming holes. Further analysis and AGCM simulations link these atmospheric circulation trends to Indo-Pacific SST warming. In response to the Indo-Pacific SST warming, tropical tropospheric warming leads to the equatorward shift of the mid-latitude jet stream, and subsequently, anomalous easterly and high pressure over mid-high latitudes tend to influence the Siberia High and Aleutian low, resulting in the LSAT cooling over the CEE and NWNA.
Article
Full-text available
Maize is one of the important food crops in the Czech Republic, its growth and productivity are influenced by climate change. This study investigated the influence of precipitation under recent climate change on maize yield both for grain and silage in the whole Czech Republic during 2002–2019. Total maize yield and yield rate increased in the Czech Republic from 1961 to 2010, but they became to decrease after 2010. This is in line with the tendency of decreased precipitation and an increase in temperature after 2010, and changes are especially significant during the maize growing period, which indicates the importance of temperature and precipitation. In detail, there is a low to moderate negative correlation (−0.39 to −0.51) between grain maize yield and the average temperature in August for almost all the regions. While there is a low negative correlation between silage maize yield with the average temperature in July and August from some regions. The precipitation in July exhibited moderate to high positive correlation (0.54–0.79) to grain maize yield rate for almost all the regions, and it had low to moderate positive correlation (0.35–0.70) to silage maize yield rate for all the regions. Water deficit exhibited a negative correlation with both maize yield rate and its influence mainly in July for silage but both in July and August for grain. Farmer’s profit from grain maize is influence by yield rate, temperature, precipitation, and water deficit. A positive correlation was found between profit and grain yield rate and precipitation from July and August, while a negative correlation was detected between profit and water deficit and the average temperature in July and August. In conclusion, our results pointed out the factors influencing maize yield rate under changing climate conditions in the Czech Republic, and it warrants further studies on how to maintain maize production in a changing climate.
Article
Full-text available
Understanding the climatic drivers of present-day agricultural yields is critical for prioritizing adaptation strategies to climate change. However, unpacking the contribution of different environmental stressors remains elusive in large-scale observational settings in part because of the lack of an extensive long-term network of soil moisture measurements and the common seasonal concurrence of droughts and heat waves. In this study, we link state-of-the-art land surface model data and fine-scale weather information with a long panel of county-level yields for six major US crops (1981–2017) to unpack their historical and future climatic drivers. To this end, we develop a statistical approach that flexibly characterizes the distinct intra-seasonal yield sensitivities to high-frequency fluctuations of soil moisture and temperature. In contrast with previous statistical evidence, we directly elicit an important role of water stress in explaining historical yields. However, our models project the direct effect of temperature—which we interpret as heat stress—remains the primary climatic driver of future yields under climate change.
Article
Full-text available
Future climate change impacts on humans will be determined by the convergence of evolving physical climate and socioeconomic systems. Rural areas of the northeastern United States have experienced increased temperature and precipitation extremes, especially over the past three decades, and face unique challenges due to their physical isolation, natural-resource–dependent economies, and high poverty rates. To explore the impacts of future extreme events on vulnerable, rural populations in the Northeast, we analyzed future (2046–2075) annual maximum daily maximum temperature, annual minimum daily minimum temperature, annual maximum daily precipitation, and annual minimum two-week summer precipitation for two greenhouse gas emissions scenarios using four global climate models and gridded temperature and precipitation observations. We then combined those projections with estimates of county-level population and relative income for 2060 to calculate changes in extreme person-events compared to a historical period (1976–2005), with a focus on Northeast counties that are rural (less than 250,000 people) and vulnerable (in the bottom income quartile). For all counties and lowest income quartile counties in the rural Northeast, extreme heat person-events increase nineteen-fold and twenty-nine fold, respectively, under the high greenhouse gas emissions scenario, far exceeding the decrease in extreme cold person-events. The number of extreme precipitation and drought person-events does not significantly change across the rural Northeast. For the high greenhouse gas emissions scenario, counties in the bottom two income quartiles experience the 1976–2005 once-per-year daily heat event twenty-three times per year by 2046–2075; the lower greenhouse gas emissions scenario reduces the incidence to eleven times per year. Key Words: climate change, climate impacts, extreme events, northeastern United States, vulnerability.
Article
Full-text available
Significance Over the course of the 20th century, US maize yields have improved by more than a factor of five. Whereas this trend is often attributed exclusively to technological improvements, here, we also identify contributions from improved temperatures during the growing season. More than one-quarter of the increase in crop yield since 1981 is estimated to result from trends toward overall warmer conditions, but with cooling of the hottest growing-season temperatures, and from adjustments in crop timing toward earlier planting and longer maturation varieties.
Article
Full-text available
We present a novel approach to characterize the spatiotemporal evolution of regional cooling across the eastern U.S. (commonly called the U.S. warming hole), by defining a spatially explicit boundary around the region of most persistent cooling. The warming hole emerges after a regime shift in 1958 where annual maximum (Tmax) and minimum (Tmin) temperatures decreased by 0.46°C and 0.83°C respectively. The annual warming hole consists of two distinct seasonal modes, one located in the southeastern U.S. during winter and spring and the other in the midwestern U.S. during summer and autumn. A correlation analysis indicates that the seasonal modes also vary in causation. Winter temperatures in the warming hole are significantly correlated with the Meridional Circulation Index (MCI), North Atlantic Oscillation (NAO), and Pacific Decadal Oscillation (PDO). However, the variability of ocean-atmosphere circulation modes is insufficient to explain the summer temperature patterns of the warming hole.
Article
Full-text available
Sustainable management of agricultural water resources requires improved understanding of irrigation patterns in space and time. We produced annual, high resolution (30 m) irrigation maps for 1999-2016 by combining all available Landsat satellite imagery with climate and soil covariables in Google Earth Engine. Random forest classification had accuracies from 92-100% and generally agreed with county statistics (r2 = 0.88-0.96). Two novel indices which integrate plant greenness and moisture information show promise for improving satellite classification of irrigation. We found considerable interannual variability in irrigation location and extent, including a near doubling between 2002 and 2016. Statistical modeling suggested precipitation and commodity price influenced irrigated extent through time. High prices incentivized expansion to increase crop yield and profit, but dry years required greater irrigation intensity, thus reducing area in this supply-limited region. Datasets produced with this approach can improve water sustainability by providing consistent, spatially explicit tracking of irrigation dynamics over time.
Book
Full-text available
Statistical downscaling and bias correction are becoming standard tools in climate impact studies. This book provides a comprehensive reference to widely used approaches, and additionally covers the relevant user context and technical background, as well as a synthesis and guidelines for practitioners. It presents the main approaches including statistical downscaling, bias correction and weather generators, along with their underlying assumptions, skill and limitations. Relevant background information on user needs and observational and climate model uncertainties is complemented by concise introductions to the most important concepts in statistical and dynamical modelling. A substantial part is dedicated to the evaluation of regional climate projections and their value in different user contexts. Detailed guidelines for the application of downscaling and the use of downscaled information in practice complete the volume. Its modular approach makes the book accessible for developers and practitioners, graduate students and experienced researchers, as well as impact modellers and decision makers.
Article
Full-text available
Variability in meteorological patterns presents significant challenges to crop production consistency and yield stability. Meteorological influences on corn and soybean grain yields were analyzed over an 18-year period at a long-term experiment in Beltsville, Maryland, U.S.A., comparing conventional and organic management systems. Precipitation and temperature variables explained much of the yield variability, with precipitation and heat stress during the late vegetative and early reproductive phases of crop growth accounting for the majority of yield variability in all crops and management systems. Crop yields under conventional and organic management followed similar periodic patterns, but yields were 31% and 20% lower in organic than conventional corn and soybean, respectively. The efficiency of grain yield per unit precipitation was higher under conventional than organic management, highlighting the importance of crop management for optimizing production in response to meteorological variability. Periodic yield and precipitation patterns did not consistently align with global meteorological cycles such as the El Niño Southern Oscillation.
Article
Agricultural development is among the most significant forms of land-use change globally. In central North America it has consisted of cropland expansion in the early 1900s, yield intensification starting in the 1930s, and the development of large irrigated areas beginning in the 1950s. The area of this study encompasses the Midwest and Great Plains of the United States not only because significant agricultural change has occurred here but also because of the significant cooling (warming hole) there in the midcentury. This study investigates the relative contribution of agricultural development and greenhouse gas (GHG) emissions on the observed patterns of regional changes in summer temperature, precipitation, and evapotranspiration using a long-term twentieth-century reanalysis dataset (CERA-20C) as boundary conditions for simulations with the MIT Regional Climate Model (MRCM). Temperatures in the Great Plains (33°–43°N, 95°–109°W) and the Midwest (38°–48°N, 82°–109°W) would have been significantly higher in the second half of the twentieth century without the influence of agricultural development, largely due to an increase in evaporative cooling. The simulations of precipitation changes reflect a significant influence of global SST teleconnections at decadal time scales. Numerical simulations also demonstrate the competing effects of cropland expansion and yield intensification on shaping the observed pattern of increases in precipitation. Ultimately, a combination of agricultural development and decadal variability of global sea surface temperatures (SST) explains most of the observed variability of summer temperature and precipitation during the twentieth century over central North America.
Article
Both land-use changes and greenhouse gas (GHG) emissions have significantly modified regional climate over the last century. In the central United States, for example, observational data indicate that rainfall increased, surface air temperature decreased, and surface humidity increased during the summer over the course of the 20th century concurrently with increases in both agricultural production and global GHG emissions. However, the relative contributions of each of these forcings to the observed regional changes remain unclear. Results of both regional climate model simulations and observational analyses suggest that much of the observed rainfall increase – as well as the decrease in temperature and increase in humidity – is attributable to agricultural intensification in the central United States, with natural variability and GHG emissions playing secondary roles. Thus, we conclude that 20th-century land-use changes contributed more to forcing observed regional climate change during the summer in the central United States than increasing GHG emissions.
Article
1. Average annual temperature over the contiguous United States has increased by 1.2°F (0.7°C) for the period 1986–2016 relative to 1901–1960 and by 1.8°F (1.0°C) based on a linear regression for the period 1895–2016 (very high confidence). Surface and satellite data are consistent in their depiction of rapid warming since 1979 (high confidence). Paleo-temperature evidence shows that recent decades are the warmest of the past 1,500 years (medium confidence). 2. There have been marked changes in temperature extremes across the contiguous United States. The frequency of cold waves has decreased since the early 1900s, and the frequency of heat waves has increased since the mid-1960s (the Dust Bowl remains the peak period for extreme heat). The number of high temperature records set in the past two decades far exceeds the number of low temperature records. (Very high confidence) 3. Average annual temperature over the contiguous United States is projected to rise (very high confidence). Increases of about 2.5°F (1.4°C) are projected for the next few decades in all emission scenarios, implying recent record-setting years may be “common” in the near future (high confidence). Much larger rises are projected by late century: 2.8°–7.3°F (1.6°–4.1°C) in a lower emissions scenario (RCP4.5) and 5.8°–11.9°F (3.2°–6.6°C) in a higher emissions scenario (RCP8.5) (high confidence). 4. Extreme temperatures in the contiguous United States are projected to increase even more than average temperatures. The temperatures of extremely cold days and extremely warm days are both expected to increase. Cold waves are projected to become less intense while heat waves will become more intense. The number of days below freezing is projected to decline while the number above 90°F will rise. (Very high confidence)