Field Crops Research 176 (2015) 110

Contents lists available at ScienceDirect

Field Crops Research

journal homepage: www.elsevier.com/locate/fcr

Differences between observed and calculated solar radiations and

their impact on simulated crop yields
Jing Wang a, , Enli Wang b , Hong Yin c , Liping Feng a , Yanxia Zhao d
a

College of Resources and Environmental Sciences, China Agricultural University, Beijing 100193, China
CSIRO Land and Water, GPO Box 1666, Canberra ACT2601, Australia
National Climate Center, China Meteorological Administration, Beijing 100081, China
d
Chinese Academy of Meteorological Sciences, Beijing 100081, China
b
c

a r t i c l e

i n f o

Article history:
Received 19 July 2014
Received in revised form 27 October 2014
Accepted 16 February 2015
Available online 9 March 2015
Keywords:
Global solar radiation
ngstrmPrescott model
JohnsonWoodward model
Crop modeling
Simulated yield potential
APSIM

a b s t r a c t
Global solar radiation, a key input variable for crop growth models, is often estimated from sunshine
duration with the ngstrmPrescott model due to limited observational data. Few studies have explored
the possible difference between the calculated and observed solar radiation data, and how this difference
transfers to differences in simulated crop yield. This paper compared observed global solar radiation and
calculated values using the ngstrmPrescott (AP) model and the JohnsonWoodward (JW) model at
different time scales. We further used the farming systems model APSIM to investigate the difference in
simulated wheat and maize yields caused by the use of observed and calculated solar radiations at eight
sites in the North China Plain (NCP). The results revealed signicant differences between the observed
and calculated solar radiations, which also varied among sites. Overall, the calculated daily global solar
radiations with the AP model and the JW model could explain 8792% of the variations in the observed
ones. The AP model performed slightly better (RMSE of 2.062.80 MJ m2 day1 , RRMSE of 14.4921.50%)
than the JW model (RMSE of 2.292.97 MJ m2 day1 , RRMSE of 16.3522.82%). Using different coefcients
(a and b) in the AP model between seasons and sites did not improve the estimated global solar radiation
compared to using xed coefcients across the study sites. In general, the calculated solar radiation in
the maize growing seasons better correlated with the observed solar radiation than those in the wheat
seasons at most of the sites. As a consequence, the simulated potential yields of maize using the two
sources of solar radiation data were much more closely correlated than those of wheat. While both the
calculated and observed solar radiations showed the same directions of change with years, the rate of
change per year differed signicantly at some locations. Similar results were also found in the trend
of change in simulated wheat and maize potential yields using the two sources of solar radiations, i.e.,
similar change direction, but different rates of change. These results indicate that further research is
needed in order to construct more reliable solar radiation data for use in the crop modeling. This will
include stricter quality control of observed data and use of improved calculation method for global solar
radiation.
Crown Copyright 2015 Published by Elsevier B.V. All rights reserved.

1. Introduction
Solar radiation is a key input variable for crop growth models. Unfortunately, it is often not available at many locations,
and has to be estimated from more frequently measured sunshine hours or air temperature (ngstrm, 1924; Prescott, 1940;
Bristow and Campbell, 1984; Hoogenboom, 2000; Rivington et al.,
2005). Inaccuracy in the estimation of global solar radiation could
impact signicantly on the simulated crop yields (Rivington et al.,

Corresponding author. Tel.: +86 10 6273 4636; fax: +86 10 6273 4636.
E-mail address: wangj@cau.edu.cn (J. Wang).
http://dx.doi.org/10.1016/j.fcr.2015.02.014
0378-4290/Crown Copyright 2015 Published by Elsevier B.V. All rights reserved.

2006). Nowadays, crop models have been increasingly used for

estimation of crop potential yield, for determining crop yield gap,
rening management strategies and quantifying the impacts of
climate change on crop production (Lobell et al., 2009; Wang
et al., 2012, 2014). In spite of the wide use of solar radiation data estimated using different methods as input for crop
modeling, the potential inaccuracy of the estimations and the
subsequent impact on simulated crop yield has been rarely investigated.
Several studies showed that sunshine duration-based estimation models perform better than air temperature-based estimation
models (Chen et al., 2004; Podesta et al., 2004; Rivington et al.,
2005). The ngstrmPrescott (AP) model has been widely used

J. Wang et al. / Field Crops Research 176 (2015) 110

to calculate daily global solar radiation from sunshine duration

(Doorenbos and Pruitt, 1977; Allen et al., 1998; Almorox and
Hontoria, 2004), and the calculated solar radiation values are subsequently used in crop models. A large number of studies compared
the difference between observed global solar radiation and those
calculated with the ngstrmPrescott model (Boisvert et al., 1990;
Iziomon and Mayer, 2002; Liu et al., 2009a,b). A general conclusion
was that the ngstrmPrescott model could effectively estimate
daily global solar radiation, but its coefcients a and b depend on
sites and seasons (Podesta et al., 2004; Tymvios et al., 2005), though
a single set of coefcient was often used in simulation studies to
analysis crop potential yield and yield gaps in a given region (Liu
et al., 2010, 2012). Another method to convert sunshine duration to
solar radiation is the JohnsonWoodward (JW) model. It contains
only one empirical constant (Johnson et al., 1995; Woodward et al.,
2001; Rivington et al., 2005). It has been used effectively in tropical rainforest, New Zealand and UK, but has not been evaluated in
other climate regions.
One study (Pohlert, 2004) compared simulated maize yield
under the potential and water-limited production conditions using
observed daily global solar radiation and daily global solar radiation
estimated with three empirical solar radiation models (ngstrm,
Bristow and Campbell, the Allen global solar radiation models) at
two locations in temperate region (Wageningen in the Netherlands

and Crdoba in Argentina) and one location in the tropics (Los Banos
in the Philippines). The results showed that all the three models
could be applied to close incomplete global solar radiation series
for maize growth simulation with the WOFOST model at temperate
locations. However, the Bristow and Campbell, and the Allen global
solar radiation models may not be used to generate daily data for a
full season at tropical locations, as the simulated yield distributions
differ signicantly due to poor prediction skill of the solar radiation
models. This implies that the accuracy of solar radiation estimation
changes with the methods used and also across regions, potentially
leading to changed spatiotemporal distribution of simulated crop
yield.
In the North China Plain, one of the most important agricultural production regions in China, the solar radiation environment
has been changing since the 1980s due to increasing atmospheric
aerosol caused by industrial development. There has been no
investigation on the accuracy of solar radiation data estimated
using different methods and their impact on simulated crop yield,
although many modeling studies have been conducted using the
solar radiation data estimated with the AP model (Chen et al.,
2010b,c; Wang et al., 2012, 2014). It is unknown whether the trend
of change in simulated crop yields remains consistent if estimated
versus measured solar radiation data are used in the modeling. This

Table 1
Periods of available observed solar radiation data and the data used in this study,
and the ratio of missing to total number of observed data for the used records at the
eight study sites in the North China Plain.
Sites

Available
records

Used
records

Ratio of missing data to total

number of used records

Beijing
Leting
Tianjin
Jinan
Juxian
Zhengzhou
Nanyang
Gushi

19572011
19922011
19592011
19612011
19902011
19612011
19902011
19612011

19612011
19932011
19612011
19612011
19902011
19612011
19922010
19612011

0.04%
0.4%
0.8%
0.2%
0.6%
0.7%
0.4%
0.07%

is particularly important for climate change impact studies, where

accurate quantication of the trends of change in both climatic
variables and simulated crop yield are essential.
The objectives of this study are to investigate: (1) the performance of the ngstrmPrescott model and the Johnson
Woodward model to estimate global solar radiation at different
time scales in the North China Plain, (2) if the performance of
the ngstrmPrescott model can be improved by using different
coefcients between seasons and sites, (3) the difference between
the change trends in observed and calculated global solar radiation
at the eight sites in the North China Plain, and (4) whether the simulated yield time series with estimated solar radiation data can be
used for trend analysis in climate change impact studies.
2. Materials and methods
2.1. Study sites and climate data
Eight sites were selected in this study, i.e., Beijing (39.8 N,
116.47 E, 31 m), Leting (39.43 N, 118.88 E, 11 m), Tianjin
(39.08 N, 117.07 E, 3 m), Jinan (36.68 N, 116.98 E, 52 m), Juxian (35.58 N, 118.08 E, 107 m), Zhengzhou (34.72 N, 113.65 E,
110 m), Nanyang (33.03 N, 112.58 E, 129 m) and Gushi (32.02 N,
115.07 E, 57 m). These sites are selected because observed solar
radiation data are available on site. They are roughly uniformly
distributed across North China Plain (Fig. 1). Annual average air
temperatures at Beijing, Leting, Tianjin, Jinan, Juxian, Zhengzhou,
Nanyang and Gushi were 12.7, 12.1, 13.2, 15, 13.4, 15, 16 and
16.1 C, while annual precipitation totals were 553, 546, 536, 702,
803, 644, 801 and 1067 mm, respectively.
Table 1 shows the periods of available observed records of solar
radiation, data used in this study and the ratio of missing data to
total number of observed data using at the eight study sites. The

Fig. 1. The North China Plain (NCP) and the locations of the eight study sites.

J. Wang et al. / Field Crops Research 176 (2015) 110

solar radiation data have been quality-controlled by China Meteorological Administration (Ma et al., 1998). For this study, we only
used part of the observed records to ensure that the ratio of missing
data was lower than 0.8% at all the sites (Table 1). Daily sunshine
hour at the selected sites are available, together with daily average,
maximum and minimum temperatures and rainfall, from the China
Meteorological Administration.
2.2. Calculation of global solar radiation from sunshine duration

Rs =

n
a+b
N

Ra

(1)

where Rs is daily global solar radiation (MJ m2 day1 ), n is actual

sunshine duration (h), i.e., the time during the day of direct light
intensity above 120 W m2 . N is maximum possible sunshine duration (h), Ra is extraterrestrial radiation (MJ m2 day1 ), a and b are
regression constants.
Ra can be estimated from the solar constant, the solar declination
and the time of the year by:
24(60)
Gsc dr [s sin() sin() + cos() cos() sin(s )]
Ra =


where J0,s is the direct beam (J m2 s1 ) and J0,d is the diffuse component (J m2 s1 ).
J0,s is given by (Campbell, 1977):
J0,s = 1367

2p
sin ( 1/ sin  )


(2)

 = 0.64 + 0.12 cos 2

J0,d = J0,p (fblue (1 c) + fcloud c)

= 0.409 sin

 2

J 1.39

(4)

s = arccos[ tan() tan()]

(5)

365

s is given by:

N is given by:
24
N=
s


(6)

Here, we consider the following three methods to determine a

and b coefcients in the ngstrmPrescott model:
(1) AP-1: 0.18 and 0.55 were used for a and b across the NCP,
respectively (Wang et al., 2008, 2014; Chen et al., 2010c).
(2) AP-2: a and b coefcients were tted through linear regression
based on Eq. (1) for the whole study period at each of the study
sites. Consequently each site has different a and b coefcients.
(3) AP-3: a and b coefcients were tted based on Eq. (1) for different seasons (winter: DecemberFebruary, spring: MarchMay,
summer: JuneAugust, autumn: SeptemberNovember) during
the whole study period at each of the study sites. Consequently
a and b coefcients varied between sites and seasons.
2.2.2. JohnsonWoodward model
The JohnsonWoodward model was also used to calculate daily
global solar radiation from sunshine duration (Johnson et al., 1995;
Woodward et al., 2001; Rivington et al., 2005). Daily global solar
radiation Rs (MJ m2 day1 ) is given by:
Rs =

nJ0,s 3600 + NJ0,d 3600

1, 000, 000

(7)

(10)

(11)

where c is the average daily fraction of cloud cover, given by:

c =1

n
N

(12)

J0,p is the potential total clear sky mean daily irradiance:

J0,p = 1367

p
sin (1 +  1/ sin


(13)

fblue and fcloud are the relative different radiation intensities under
blue sky and cloud conditions, respectively:

fcloud = Ffblue

where J is the day of year.

is given by:

(9)

J0,d is given by:

dr = 1 + 0.033 cos

(3)

sin  = sin  sin + cos  cos

fblue =

365

J
365

 is the solar elevation at noon in degrees:

where Gsc is solar constant (0.082 MJ m2 min1 ), dr is inverse relative distance EarthSun, s is sunset hour angle, is latitude (rad),
is solar declination.dr is given by:

 2J 

(8)

where p is the fraction of radiation in full spectrum sunlight (here

1 is taken) and 1367 is the solar constant in J m2 s1 . , the atmospheric transimissivity, is given as (Woodward et al., 2001):

2.2.1. ngstrmPrescott model

The ngstrmPrescott model was used to calculate daily global
solar radiation from sunshine duration (ngstrm, 1924; Prescott,
1940; Allen et al., 1998):

1  1/ sin

1 +  1/ sin

(14)
(15)

where F is the regression constant, representing the relative intensity of diffuse solar radiation from cloudy skies.
Here, daily F values were tted using observed solar radiation
for each day per year (LADSS, 2005; Rivington et al., 2005). Then the
average of F values of all the study period was used as the optimized
F value for each site.
2.3. The agricultural system model
The Agricultural Production System sIMulator (APSIM, version
5.3) was used to simulate the potential yield of wheat and maize.
In APSIM, potential growth rate of daily biomass is a function of
light interception, radiation use efciency (RUE) and temperature.
Grain yield is calculated based on grain number, grain lling rate
and assimilate re-translocation. RUE is treated as a constant for a
given species and is stage-dependent.
APSIM was well-tested and widely used in the NCP (Wang et al.,
2007, 2012, 2013; Chen et al., 2010a,b). The results from these previous studies indicated that in general, APSIM model was able to
explain the observed crop growth and yield in response to water
and nitrogen inputs in the NCP, and simulate crop growth and
yield accurately under the potential condition (i.e. without water
or nutrient stress) (Chen et al., 2010a).
Due to limited observed solar radiation data in NCP, almost
all the previous modeling studies calculated global solar radiation from observed sunshine hours (using Eq. (1)) to simulate crop
growth. A few studies indicated that APSIM tended to underestimate maize yield under high levels of water and nitrogen inputs
(potential yield). This has been attributed to possible underestimation of radiation use efciency (RUE) under the changed radiation
conditions in the last 2030 years (Chen et al., 2010c). Questions
have also been asked if the calculated solar radiation from Eq. (1)
is appropriate to represent the actual solar radiation conditions
(Pohlert, 2004).

J. Wang et al. / Field Crops Research 176 (2015) 110

Table 2
Simulation periods and planting date of wheat and maize at the study sites.
Site

Simulation periods

Beijing
Leting
Tianjin
Jinan
Juxian
Zhengzhou
Nanyang
Gushi

19612011
19932011
19612011
19612011
19902011
19612011
19922010
19612011

Planting date
Wheat

Maize

October 3rd
October 3rd
October 3rd
October 10th
October 12th
October 13th
October 19th
October 28th

June 16th
June 16th
June 16th
June 21th
June 19th
June 5th
June 5th
June 6th

Table 3
APSIM parameterization for wheat variety (Baofeng 7228) used in the study.

of linear regression lines, and evaluated using Students t-test at

95% and 99% levels.


 n
1
RMSE = 
(Y i Xi )2
n

RRMSE =

(16)

i=1


 n

1 
1
(Y i Xi )2

Xaver

(17)

i=1

where Xi and Yi are the observed and simulated values respectively,

Xaver is the average of the observed values, and n is the number of
observations.
3. Results

Parameters

Values

Sensitivity to vernalisation
Sensitivity to photoperiod
Thermal time from beginning of grain-lling to
maturity ( C day)
Coefcient of kernel number per stem weight
at the beginning of grain-lling (g per stem)
Potential grain-lling rate (g per kernel per
day)
Phyllochron interval ( C day/leaf appearance)

3.0
3.8
490

3.1. Accuracy of calculated global solar radiations at daily,

monthly and seasonal scales

23.0

Table 5 shows tted AP coefcients a and b and JW coefcient F,

RMSE, RRMSE and R2 between calculated and observed daily global
solar radiations for the eight sites. Considering the variation in AP
coefcients between sites, the AP-2 coefcient a varied between
0.14 and 0.21 (averaged 0.18) and the AP-2 coefcient b varied
between 0.49 and 0.58 (averaged 0.54). Considering the variation
in AP coefcients between seasons and sites, the AP-3 coefcient a
varied between 0.14 and 0.23 (averaged 0.19) for winter, 0.13 and
0.19 (averaged 0.17) for spring, 0.15 and 0.21 (averaged 0.19) for
summer, 0.14 and 0.2 (averaged 0.17) for autumn, The AP-3 coefcient b varied between 0.47 and 0.53 (averaged 0.51) for winter,
0.54 and 0.59 (averaged 0.57) for spring, 0.51 and 0.56 (averaged
0.53) for summer, 0.51 and 0.57 (averaged 0.55) for autumn. However, AP-2 and AP-3 did not improve the simulated daily global solar
radiation as compared to AP-1 because there were no signicant
difference in R2 , RMSE and RRMSE between observed and calculated daily global solar radiation using the three models. Therefore,
the calculated solar radiation data from AP-1, i.e. with xed a and b
coefcients, were used to compare with those from the JW model
and observed global solar radiation.
Overall, the calculated solar radiation values with the AP-1
model and the JW model could explain 8792% of the variations in the observed ones (Table 5). The AP-1 model performed
slightly better with lower RMSE of 2.082.80 MJ m2 day1 (RRMSE
of 14.4921.50%) as compared to the JW model with RMSE of
2.292.97 MJ m2 day1 (RRMSE of 16.3522.82%). The above
results suggested that the AP-1 model and the JW model could
explain large variations in daily global solar radiation in the NCP.
However, the seasonality of the solar radiation, i.e., both calculated and observed solar radiations were low in winter and high
in summer, might have contributed to a high R2 .
Therefore, the ranges of monthly total solar radiation summed
from the observed and calculated daily values were further compared (Fig. 2). While the calculated values captured the seasonal
change in global solar radiation, they showed different distribution
(in terms of mean and percentiles among the years) from those
of the observed values in each individual month. For individual
months, the monthly total of calculated solar radiation values with
the AP-1 model and the JW model could explain 3493% of the variations in the monthly total of the observed values, with the R2 and
RRMSE varied signicantly between months and locations (Fig. 3).
The AP-1 model performed slightly better in estimating monthly
total solar radiation than the JW model for all the study sites. In
general, January, MarchJuly, and December (where MarchMay
is the period of active growth of wheat) were the period when the
calculated values matched less well to observed values considering

0.0025
85.0

Here we use APSIM to simulate the potential yield of the

wheatmaize double cropping system at the selected sites, using
both the observed and calculated global solar radiations with the
AP model and the JW model. We then compared the simulation
results and investigate if signicant difference exists in the simulations results. For simplicity, one wheat variety (Baofeng 7228)
and one maize variety (Nongda 108) were used at all the sites during the studies period. Table 2 summarized the simulation periods
and planting date of winter wheat and summer maize at the study
sites. Wheat and maize were harvested at physiological maturity.
The key varietal parameters of wheat and maize were shown in
Tables 3 and 4 (Wang et al., 2013, 2014).

2.4. Statistical analysis

Linear regression analysis was performed to analyze the accuracy of the calculated solar radiation at daily, monthly and seasonal
scales, using the slope, the intercept, the coefcient of determination (R2 ) of the regression lines, root mean square error (RMSE),
Eq. (16) and relative root square error (RRMSE), Eq. (17) between
observed and calculated data. In addition, the trend of change in
both observed and calculated global solar radiations, and in the
potential yield of wheat and maize simulated with the observed
and calculated solar radiation data were investigated, as the slope

Table 4
APSIM parameterization for maize variety (Nongda 108) used in the study.
Parameters

Values

Maximum grain numbers per head

Grain-lling rate (mg/grain/day)
Thermal time (TT) required from emergence to
end of juvenile ( C day)
Change in TT required to oral initiation per
hour photoperiod increase ( C/h)
Thermal time required from owering to
maturity ( C day)
Thermal time required from owering to
starting grain-lling ( C day)

600
10
190
10.0
720
130

J. Wang et al. / Field Crops Research 176 (2015) 110

Table 5
Fitted AP coefcients a and b and JW coefcient F, RMSE (MJ m2 day1 ), RRMSE (%) and R2 between observed daily global solar radiation and calculated daily global solar
radiation with the AP-1, AP-2, AP-3 models and the JW model for the study sites.
Site

Beijing

Leting

Tianjin

Jinan

Juxian

Zhengzhou

Nanyang

Gushi

AP-1
RMSE (RRMSE)
R2

2.08 (14.49)
0.92

2.12 (15.16)
0.91

2.55 (18.30)
0.88

2.69 (20.00)
0.88

2.16 (15.72)
0.90

2.39 (17.98)
0.89

2.37 (19.39)
0.88

2.80 (21.50)
0.88

AP-2
a
b
RMSE (RRMSE)
R2

0.17
0.56
2.07 (14.43)
0.92

0.20
0.52
2.12 (15.12)
0.91

0.19
0.53
2.55 (18.29)
0.88

0.14
0.56
2.48 (18.46)
0.89

0.21
0.51
2.11 (15.30)
0.91

0.18
0.54
2.37 (17.87)
0.89

0.20
0.49
2.37 (19.43)
0.89

0.15
0.58
2.76 (21.22)
0.88

AP-3
a
Win
Spr
Sum
Aut
b
Win
Spr
Sum
Aut
RMSE (RRMSE)
R2

0.20
0.16
0.17
0.16

0.23
0.17
0.20
0.19

0.21
0.19
0.18
0.18

0.14
0.13
0.15
0.14

0.22
0.19
0.21
0.20

0.20
0.17
0.19
0.17

0.19
0.17
0.21
0.18

0.14
0.14
0.17
0.15

0.53
0.59
0.56
0.57
2.06 (14.35)
0.92

0.47
0.57
0.54
0.54
2.08 (14.85)
0.91

0.50
0.56
0.52
0.54
2.51 (18.01)
0.88

0.54
0.58
0.54
0.56
2.69 (20.01)
0.87

0.47
0.55
0.53
0.51
2.07 (15.01)
0.91

0.52
0.56
0.51
0.55
2.36 (17.76)
0.89

0.48
0.54
0.51
0.52
2.32 (18.95)
0.89

0.58
0.59
0.55
0.57
2.75 (21.14)
0.88

JW
F
RMSE (RRMSE)
R2

0.759
2.42 (16.90)
0.92

0.605
2.29 (16.35)
0.91

0.661
2.82 (20.26)
0.87

0.479
2.95 (21.93)
0.89

0.617
2.37 (17.21)
0.90

0.595
2.75 (20.73)
0.89

0.566
2.76 (22.57)
0.88

0.597
2.97 (22.82)
0.88

Win: winter; Spr: spring; Sum: summer; Aut: autumn.

R2 and RRMSE together, and at Tianjin, Jinan and Gushi, the calculated solar radiation departed much more from the observed ones
than other places (Fig. 3).
The total solar radiation during the wheat growing season estimated from the calculated solar radiations with the AP-1 model

and the JW model accounted for only 2855% of the variations in

the seasonal total of observed solar radiation, with RMSE range of
110271 MJ m2 (RRMSE range of 3.679.31%) for the AP-1 model
and RMSE range of 122275 MJ m2 (RRMSE range of 4.149.46%)
for the JW model (Fig. 4, left panels), though the correlation

Fig. 2. Ranges of monthly total global solar radiation calculated from observed and calculated values (using the AP-1 model and the JW model). Box plots show 0th, 5th,
25th, 50th, 75th, and 100th percentiles of the values.

J. Wang et al. / Field Crops Research 176 (2015) 110

Fig. 3. Coefcient of determination (R2 ) (a) and root mean square error (RMSE) (b) between monthly totals of observed and calculated global solar radiations (using the AP-1
model and the JW model) at the eight study sites in NCP.

between the seasonal totals were highly signicant (p < 0.01). In

comparison, the total solar radiations during the maize growing season were much better correlated, with higher R2 range
of 0.420.73 for both models, a RMSE range of 103227 MJ m2
(RRMSE range of 4.8811.38%) for the AP-1 model and a RMSE range
of 95271 MJ m2 (RRMSE range of 4.513.54%) for the JW model
(Fig. 4, right panels). The correlations were the best at Beijing for
total solar radiation in both wheat and maize seasons and worst for
total solar radiation at Zhengzhou in wheat season and at Tianjin
in maize season.
3.2. Correlation between simulated crop potential yields using
the two sources of solar radiation data
At all the eight sites, the simulated wheat potential yields with
observed and calculated solar radiation data were signicantly

correlated (p < 0.01). However, there were great variations in

R2 and RRMSE between sites, with R2 (RRMSE) ranging from
0.29 (14.45%) for the AP-1 model and 0.27 (14.25%) for the JW
model at Gushi to 0.61 (6.42%) at Beijing for the AP-1 model
and 0.62 (9.83%) for the JW model. For maize, the simulated
potential yields with the two sources of solar radiation data
were more closely correlated than those of wheat at six of
the eight sites except for Tianjin and Jinan (Fig. 5, right panels). The simulated yield with calculated solar radiation could
explain more than 69% of the variation in maize yield simulated with the observed solar radiation at ve of the eight sites
(Beijing, Leting, Juxian, Nanyang and Gushi), and 3152% of the
variation at the other three sites. However, RRMSE between
the simulated potential yields with the two sources of solar
radiation were higher for maize than wheat at all the study
sites.

Fig. 4. Coefcient of determination (R2 ) and root mean square error (RMSE) between total solar radiation (observed versus calculated global solar radiations using the AP-1
model and the JW model) in the wheat (left panels) and maize (right panels) growing seasons at the eight study sites in NCP. All correlations are at p < 0.01 signicance level.
BJ: Beijing; LT: Leting; TJ: Tianjin; JN: Jinan; JX: Juxian; ZZ: Zhengzhou; NY: Nanyang; GS: Gushi.

J. Wang et al. / Field Crops Research 176 (2015) 110

Fig. 5. Coefcient of determination (R2 ) and root mean square error (RMSE) between simulated wheat (left panels) and maize (right panels) yields using observed and
calculated solar radiation data (using the AP-1 model and the JW model) at the eight study sites in NCP. All correlations are at p < 0.01 signicance level. BJ: Beijing; LT: Leting;
TJ: Tianjin; JN: Jinan; JX: Juxian; ZZ: Zhengzhou; NY: Nanyang; GS: Gushi.

3.3. Trends of change in observed and calculated values of global

solar radiations and simulated crop potential yields
Fig. 6 shows the trends in observed and calculated total global
solar radiation during the growing seasons of wheat (OctoberMay)
and maize (JuneSeptember) at the eight sites in the NCP. For the
wheat season, total solar radiations from both the observed and
calculated daily solar radiations show a declining trend with time
except for Juxian and Nanyang (Fig. 6, left panels). However, at Beijing and Gushi, observed values were higher before 1980 but then
lower than calculated values thereafter, leading to smaller rates
of decline in the calculated seasonal total solar radiation. At Jinan,
calculated values with the AP-1 model were higher than observed
values in most of the years, but had a very similar declining trend
to that of the observed values. At Zhengzhou, there was a higher
rate of decline in the calculated seasonal total solar radiation than
observed ones. At other sites, there were no systematic deviations
of calculated solar radiations from the observed totals.
For the maize season, similar to the wheat season, the total seasonal solar radiations showed a declining trend for all the sites
(p < 0.05). At Beijing, calculated values with the JW model were
higher than observed values in most of the years though they had
a very similar declining trend to that of the observed values. At
Tianjin, Jinan and Zhengzhou, the calculated solar radiation was
higher than the observed ones before 2000, and they became similar thereafter, leading to a higher rate of decline in the calculated
solar radiations. Juxian had the opposite situation, with a sharper
decline in the observed solar radiations. At Nanyang, the calculated
solar radiation was higher than the observed ones before 2000, and
the observed ones became higher thereafter, leading to a sharper
rate of decline in the calculated solar radiations. At the other two
sites (Leting and Gushi), there were no systematic deviations of
calculated solar radiations from the observed totals.
The trends in simulated potential yield of wheat and maize
(Fig. 7) were consistent with those in the seasonal solar radiations
(Fig. 6). This is obvious because at potential production level, crop
growth was not limited by water and nutrient stresses, and the
yield was signicantly correlated with crop season total solar radiation (p < 0.01). The results also revealed that the trend of change

in simulated crop yield with calculated solar radiation could be

signicantly different from the trend in simulated yields with
observed solar radiation data, though both could reect the declining trend. The difference between the trends of change in the
simulated crop yield varied between sites, and between wheat and
maize.
4. Discussion
Due to unavailability of measured solar radiation data, crop
modeling studies have been often conducted using solar radiation
values calculated from sunshine hours with the ngstrmPrescott
model. This study compared the observed global solar radiation
and calculated global solar radiation with the ngstrmPrescott
(AP) model and the JohnsonWoodward (JW) model, at different
time scales. The results revealed signicant differences between
the observed and calculated solar radiations in the North China
Plain, which also varied among sites. In general, the calculated
solar radiation in the maize growing seasons better correlated with
the observed solar radiation than those in the wheat seasons. As
a consequence, the potential yields of maize simulated using the
two sources of solar radiation data (observed and calculated) were
much more closely correlated than those of wheat. While both the
calculated and observed solar radiations showed the same directions of change with years, the rate of change per year differed
signicantly at some locations. Similar results were also found in
the trend of change in wheat and maize potential yields simulated
using the two sources of solar radiations, i.e., similar change direction, but different rates of change. Rivington et al. (2006) evaluated
the uncertainty in simulated crop yield resulted from different
sources of solar radiation data in the UK, and found the JW model
performed better than temperature-based estimation models overall but not at all the study sites and for all yield assessment metrics.
All solar radiation estimation models produced over- or underestimation yield errors of >1t ha1 in at least one site in the UK.
These results indicate that further research is needed in order to
construct more reliable solar radiation data for crop modeling.
For solar radiation records, the absolute accuracy of observed
global solar radiation measured by the pyranometer in China may

J. Wang et al. / Field Crops Research 176 (2015) 110

Fig. 6. Trends in observed and calculated averaged global solar radiations (using the AP-1 model and the JW model) during growing season of wheat (OctoberMay, left
panels) and maize (JuneSeptember, right panels) at eight sites in the NCP. **Signicant at p < 0.01; *signicant at p < 0.05. The black line is the regression line of observed
values with years. The red line is the regression line of calculated values using the AP-1 model with years. The blue line is the regression line of calculated values using the
JW model with years. (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

also be debatable even calibration and maintenance procedures

have been standardized (Che et al., 2005). Possible observational
errors were caused by the polluted camera lens, the temperature
error, the consine and the tilt error, etc. Previous studies have
showed that strict quality control was needed for ground solar radiation data in China, and the quality assessment process could have
signicant effects on the data from individual solar radiation stations (Shi et al., 2008). Tang et al. (2011) revisited the solar radiation
data by combining quality-controlled observed data and two solar
radiation models, and found that their new trend of solar radiation
decline was less in magnitude than trend slopes estimated in previous studies. This highlights the importance of quality control on
the observed data.
Previous studies indicated that the ngstrmPrescott
coefcients a and b had to be varied across regions and with

seasons in order to get accurate estimation of solar radiation

(Boisvert et al., 1990; Podesta et al., 2004; Liu and Li, 2007;
Liu et al., 2009a,b). Our study showed that using different AP
coefcients a and b between seasons and sites in the AP model did
not improve simulated daily global solar radiation compared to
xed AP coefcients of a = 0.18 and b = 0.55 as recommended for
temperate and cold zones by Doorenbos and Pruitt (1977). This
proves the spatial applicability of AP model and supports its spatial
interpolation with xed coefcients in the NCP. Moreover, it also
supports previous work using a single set of coefcient in the NCP
for simulation studies to analysis crop potential yield and yield
gaps (Wang et al., 2008, 2012, 2013, 2014; Chen et al., 2010c). The
JW model could also explain large part of the variations in observed
daily global solar radiation. However, the JW coefcient F varied
with sites from 0.479 at Jinan to 0.759 at Beijing, which are close to

J. Wang et al. / Field Crops Research 176 (2015) 110

Fig. 7. Trends in simulated potential yield of wheat (left panels) and maize (right panels) using observed and calculated global solar radiations (using the AP-1 model and
the JW model) at eight sites in the NCP. **Signicant at p < 0.01; *signicant at p < 0.05. The black line is the regression line of simulated potential yield using observed global
solar radiation with years. The red line is the regression line of simulated potential yield using calculated global solar radiation (the AP1-model) with years. The blue line is
the regression line of simulated potential yield using calculated global solar radiation (the JW model) with years. (For interpretation of the references to color in this gure
legend, the reader is referred to the web version of this article.)

mean F in the UK (Rivington et al., 2005; Miller et al., 2008). Therefore, the JW coefcients F need to be retted when applied in other
sites in the NCP. In addition, a few studies have proposed improved
ways of estimating solar radiation data by including other location
and climate variables in addition to sunshine hours and showed
better estimations than those from the simple ngstrm equation
(Yang et al., 2006; Tang et al., 2010). These new approaches should
be tested in future crop modeling studies, together with the quality
control of the limited solar radiation observations.
Both of observed and calculated global solar radiations showed
a decreasing trend at the study sites in the NCP. This phenomenon
has been found also in other climate regions in China (Che et al.,
2005; Wang et al., 2009; Ma et al., 2012). A large number of studies attributed this phenomenon to the decline in sunshine duration

due to increasing emissions of anthropogenic aerosols in China (Che

et al., 2005; Xia et al., 2006). Zhang et al. (2004) and Ma et al.
(2012) found there were other possible factors besides sunshine
duration resulting in the decrease in global solar radiation. As a
result of solar radiation decline, simulated crop potential yields
decreased signicantly with years, which was consistent with the
results of Chen et al. (2010c). However, decline in total solar radiation due to increased atmospheric aerosol have increased the
fraction of diffuse radiation, which may lead to increase radiation use efciency of the crops. This may compensate some of the
negative effects of the decline in global solar radiation on crop
growth (Chen et al., 2010c), and may need to be further considered in crop models that use total solar radiation as inputs like
APSIM.

10

J. Wang et al. / Field Crops Research 176 (2015) 110

5. Conclusion
While solar radiation is one of the main dening factors for
crop growth in crop simulation modeling, it is often estimated
due to unavailability of measurement data. Results of this study
revealed signicant differences between the observed solar radiation and those calculated using the widely applied methods, and
these differences also varied among sites. In North China Plain,
ngstrmPrescott (AP) model performed best in calculation of
global solar radiation and could explain 8792% of the variations
in the observed ones. Using different coefcients (a and b) in the
AP model between seasons and sites did not improve the estimated
global solar radiation. In general, the calculated solar radiation in
the maize growing seasons better correlated with the observed
solar radiation than those in the wheat seasons at most of the sites.
As a consequence, the simulated potential yields of maize using the
two sources of solar radiation data were much more closely correlated than those of wheat. While both the calculated and observed
solar radiations showed the same directions of change with years,
the rate of change per year differed signicantly at some locations.
Similar results were also found in the trend of change in simulated
wheat and maize potential yields using the two sources of solar
radiations. These results indicate that further research is needed in
order to construct more reliable solar radiation data for use in the
crop models.
Acknowledgements
This work is supported by National Science Foundation of
China (41101046) and National Basic Research Program of China
(2013CB430205). We would like to thank China Meteorological
Administration for providing the historical climate data and agrometeorological data. The authors acknowledge the anonymous
referees for their valuable comments.
References
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., FAO Irrigation & Drainage Papers 56 1998.
Crop Evapotranspiration: Guidelines for Computing Crop Water Requirements.
FAO.
Almorox, J., Hontoria, C., 2004. Global solar radiation estimation using sunshine
duration in Spain. Energy Convers. Manage. 45, 15291535.
ngstrm, A., 1924. Solar and terrestrial radiation. Q. J. R. Meteorol. Soc. 50, 121125.
Boisvert, J.B., Hayhoe, H.N., Dube, P.A., 1990. Improving the estimation of global solar
radiation across Canada. Agric. Forest Meteorol. 52, 275286.
Bristow, K.L., Campbell, G.S., 1984. On the relationship between incoming solar radiation and daily maximum and minimum temperature. Agric. Forest Meteorol.
31, 159166.
Campbell, G.S., 1977. An introduction to environmental biophysics. Springer, New
York.
Che, H.Z., Shi, G.Y., Zhang, X.Y., Arimoto, R., Zhao, J.Q., Xu, L., Wang, B., Chen, Z.H.,
2005. Analysis of 40 years of solar radiation data from China 19612000. Geophys. Res. Lett. 32, 15.
Chen, C., Wang, E.L., Yu, Q., 2010a. Modeling wheat and maize productivity as
affected by climate variation and irrigation supply in North China Plain. Agron.
J. 102, 10371049.
Chen, C., Wang, E.L., Yu, Q., 2010b. Modelling the effects of climate variability and
water management on crop water productivity and water balance in the North
China Plain. Agric. Water Manage. 97, 11751184.
Chen, C., Wang, E., Yu, Q., Zhang, Y., 2010c. Quantifying the effects of climate trends
in the past 43 years (19612003) on crop growth and water demand in the North
China Plain. Clim. Change 100, 559578.
Chen, R.S., Ersi, K., Yang, J.P., Lu, S.H., Zhao, W.Z., 2004. Validation of ve global
radiation models with measured daily data in China. Energy Convers. Manage.
45, 17591769.
Doorenbos, J., Pruitt, W.O., FAO Irrigation and drainage paper 24, Rome 1977. Guidelines for Predicting Crop Water Requirements.
Hoogenboom, G., 2000. Contribution of agrometeorology to the simulation of crop
production and its applications. Agric. Forest Meteorol. 103, 137157.
Iziomon, M.G., Mayer, H., 2002. Assessment of some global solar radiation parameterizations. J. Atmos. Solar-Terr. Phys. 64, 16311643.

Johnson, I.R., Riha, S.J., Wilks, D.S., 1995. Modelling daily net canopy photosynthesis
and its adaptation to irradiance and atmospheric CO2 concentration. Agric. Syst.
50, 135.
LADSS, 2005. Solar Radiation Fixed Factor (F). http://www.macaulay. ac.uk/LADSS/
reference.shtml
Liu, X., Li, Y., 2007. Effect of various ngstrmPrescott coefcients on reference crop
evapotranspiration a case study for the middle reach of Yellow River. Chin. J.
Agrometeorol. 28, 2935 (in Chinese).
Liu, X.Y., Mei, X.Y., Li, Y.Z., Wang, Q.S., Zhang, Y.Q., Porter, J.R., 2009a. Variation
in reference crop evapotranspiration caused by the ngstrmPrescott coefcient: locally calibrated versus the FAO recommended. Agric. Water Manage.
96, 11371145.
Liu, X.Y., Mei, X.R., Li, Y.Z., Zhang, Y.Q., Wang, Q.S., Jensen, J.R., Porter, J.R., 2009b. Calibration of the ngstrmPrescott coefcients (a, b) under different timescales
and their impacts in estimating global solar radiation in the Yellow River basin.
Agric. Forest Meteorol. 149, 697710.
Liu, Y., Wang, E., Yang, X., Wang, J., 2010. Contributions of climatic and crop varietal
changes to crop production in the North China Plain, since 1980. Global Change
Biol. 16, 22872299.
Liu, Z., Yang, X., Hubbard, K., Lin, X., 2012. Maize potential yields and yields gaps in
the changing climate of northeast China. Global Change Biol. 18, 34413454.
Lobell, D.B., Cassman, K.G., Field, C.B., 2009. Crop yield gaps: their importance, magnitudes, and causes. Annu. Rev. Environ. Resour. 34, 179204.
Ma, Y., Liu, X., Xu, S., 1998. Solar radiation dataset and its quality inspection in China.
Meteorol. Sci. Technol. 2, 5356 (in Chinese).
Ma, J., Luo, Y., Shen, Y., Liang, H., Li, S., 2012. Regional long-term trend of ground solar
radiation in China over the past 50 years. Sci. China Earth Sci. 42, 15971608.
Miller, D.G., Rivington, M., Matthews, K.B., Buchan, K., Bellocchi, G., 2008. Testing
the spatial applicability of the JohnsonWoodward method for estimating solar
radiation from sunshine duration data. Agric. Forest Meteorol. 148, 466480.
Podesta, G.P., Nunez, L., Villanueva, C.A., 2004. Estimating daily solar radiation in the
Argentine Pampas. Agric. Forest Meteorol. 123, 4153.
Pohlert, T., 2004. Use of empirical global radiation models for maize growth simulation. Agric. Forest Meteorol. 126, 4758.
Prescott, J.A., 1940. Evaporation from a water surface in relation to solar radiation.
Trans. R. Soc. Sci. Aust. 64, 114125.
Rivington, M., Bellocchi, G., Matthews, K.B., Buchan, K., 2005. Evaluation of three
model estimations of solar radiation at 24 UK stations. Agric. Forest Meteorol.
132, 228243.
Rivington, M., Matthews, K.B., Bellocchi, G., Buchan, K., 2006. Evaluating the impacts
of meteorological data inputs on the introduced uncertainty to process-based
simulation model output. Agric. Syst. 88, 451471.
Shi, G.Y., Hayasaka, T., Ohmura, A., Chen, Z.H., Wang, B., Zhao, J.Q., Che, H.Z., Xu, L.,
2008. Data quality assessment and the long-term trend of ground solar radiation
in China. J. Appl. Meteorol. Clim. 47, 10061017.
Tang, W.J., Yang, K., He, J., Qin, J., 2010. Quality control and estimation of global solar
radiation in China. Sol. Energy 84, 466475.
Tang, W.J., Yang, K., Qin, J., Cheng, C.C., He, J., 2011. Solar radiation trend across China
in recent decades: a revisit with quality-controlled data. Atmos. Chem. Phys. 11,
393406.
Tymvios, F.S., Jacovides, C.P., Michaelides, S.C., 2005. Comparative study of
ngstrms and articial neural networks methodologies in estimating global
solar radiation. Sol. Energy 78, 752762.
Wang, E., Yu, Q., Wu, D., Xia, J., 2008. Climate, agricultural production and hydrological balance in the North China Plain. Int. J. Climatol. 28, 19591970.
Wang, J., Wang, E., Feng, L., Yin, H., Yu, W., 2013. Phenological trends of winter wheat
in response to varietal and temperature changes in the North China Plain. Field
Crop. Res. 144, 135144.
Wang, J., Wang, E., Yang, X., Zhang, F., Yin, H., 2012. Increased yield potential of
wheatmaize cropping system in the North China Plain by climate change adaptation. Clim. Change 113, 825840.
Wang, J., Wang, E., Yin, H., Feng, L., Zhang, J., 2014. Declining yield potential and
shrinking yield gaps of maize in the North China Plain. Agric. Forest Meteorol.
195196, 89101.
Wang, L., Zheng, Y., Yu, Q., Wang, E., 2007. Applicability of agricultural production
systems simulator (APSIM) in simulating the production and water use of wheatmaize continuous cropping system in North China Plain. Chin. J. Appl. Ecol. 18,
24802486 (in Chinese).
Wang, Y., Huang, Y., Zhang, W., 2009. Changes in surface solar radiation in mainland
China over the period from 1961 to 2003. Clim. Environ. Res. 14, 405413 (in
Chinese).
Woodward, S.J.R., Barker, D.J., Zyskowski, R.F., 2001. A practical model for predicting
soil water decit in New Zealand pasture. N. Z. J. Agric. Res. 44, 91109.
Xia, X., Wang, P., Chen, H., Liang, F., 2006. Analysis of downwelling surface
solar radiation in China from National Centers for Environmental Prediction
reanalysis, satellite estimates, and surface observations. J. Geophys. Res. 111,
http://dx.doi.org/10.1029/2005JD006405.
Yang, K., Koike, T., Ye, B., 2006. Improving estimation of hourly, daily and monthly
solar radiation by importing global data sets. Agric. Forest Meteorol. 137, 4355.
Zhang, Y.L., Qin, B.Q., Chen, W.M., 2004. Analysis of 40 year records of solar radiation
data in Shanghai, Nanjing and Hangzhou in Eastern China. Theor. Appl. Climatol.
78, 217227.