Energy Conversion and Management 154 (2017) 168–179

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Classification of solar radiation zones and general models for estimating the MARK
daily global solar radiation on horizontal surfaces in China
⁎
Yanfeng Liu , Yong Zhou, Dengjia Wang, Yingying Wang, Yong Li, Ying Zhu
School of Environmental and Municipal Engineering, Xi’an University of Architecture and Technology, No. 13 Yanta Road, Xi’an 710055, China

A R T I C L E I N F O A B S T R A C T

Keywords: Classification of solar radiation zones constitutes the prerequisite for the establishment of regional daily global
Two-step solar radiation zoning method solar radiation (H ) estimation general model. Current zone establishment methods are ordinary based on solar
Sunshine duration radiation observation stations (SROS) which present a sparse and non-uniform distribution. As a result, the
Regional global solar radiation estimation possibility of misclassifications of the stations occurs in cases there are no records of radiation and far away from
General model
SROS. Therefore, by using k-means cluster and Support Vector Machine-Genetic Algorithm, a two-step radiation
zoning method was proposed in this paper according to: (a) H , sunshine duration, temperature and relative
humidity from 98 SROS and (b) sunshine duration, temperature and relative humidity from 562 stations without
radiation. The method is capable to combine the SROS and the stations without radiation in the process of
classification. Thus, these misclassifications have been effectively reduced and the accuracy of each classification
has been significantly improved. Based on the method, five radiation zones have been identified. Concurrently,
four sunshine-based models were obtained for each SROS and the analysis of statistical indexes indicated that the
cubic models presented the best performances in each zone. According to the best site-specific models and
radiation zones, the general models of regional H estimation were developed by introducing the geographical
parameters, including latitude and altitude. The comparative results demonstrated that the general models
proposed in this paper had better accuracies and can represent the general models for the H estimation of
stations without radiation records in China.

1. Introduction
During the last decade, the rapid economic development and steady
improvement of living standards increased rapidly China's energy
consumption. In 2012, China’s total primary energy consumption
reached to 3.62 billion coal equivalents, which equals to 1.5 times as
much as that in 2000 [1]. Energy has become a significant problem that
affects China’s economy and environment [2]. Solar energy, re-
presenting a safe, economic, environmental and renewable energy form
[3], is regarded as the alternative energy to improve the energy struc-
ture of China. Solar radiation (H ) data correspond to the basic data of
various solar energy utilization technologies [4]. Compared to tem-
perature and sunshine duration (S ) parameters, the distribution of SROS
is generally much sparser and non-uniform, resulting in the lack of H in
most of stations [5,6]. Thus, it is difficult to meet the requirement of
solar energy utilization data. At present, there are 756 surface me-
teorological observation stations (SMOS), where the weather data (in-
cluding daily dry-bulb temperature, relative humidity, wind velocity,
pressure and rainfall) are provided by China meteorological stations.
Only 122 of them measured and recorded H [7]. Therefore, it is par-
ticularly critical and essential to seek for a method to accurately esti-
mate the daily H on horizontal surfaces in China.
The recent research has ordinary adopted indirect statistical
methods to estimate the H ; that is, according to the correlation between
H and other meteorological parameters, such as S , temperature, cloud,
rainfall and fog, empirical formulas were established [8], including
sunshine [9–13], temperature [14–18] and multi-parameter [19–23]
based models. Nevertheless, the majority of these models are site-de-
pendent and it is questionable if an application to other stations without
records of H , can occur. Therefore, to estimate the daily H of these
stations without H in different areas, it is necessary to establish the
general models of regional daily H estimation. Zhou et al. [24], in-
troduced the geographical parameters to establish the general models
for estimating the monthly mean daily H , by using the data from
mainland city of China. But the models were useless in estimating the
daily H . Liu et al. [25], established two general models for estimating
the daily H in China, based on temperature and S data. The tempera-
ture-based general models proved to be less accurate, while during the

⁎
Corresponding author.
E-mail address: lyfxjd@163.com (Y. Liu).

http://dx.doi.org/10.1016/j.enconman.2017.10.043
Received 24 July 2017; Received in revised form 29 September 2017; Accepted 15 October 2017
0196-8904/ © 2017 Elsevier Ltd. All rights reserved.
Y. Liu et al.
Nomenclature
Symbols
b scalar
C custom penalty parameter
h altitude of stations (m)
H global solar radiation (MJ/m2 d)
H0 extraterrestrial horizontal radiation (MJ/m2 d)
K (x i,yi ) Kernel function
l number of training data set
S sunshine duration (h)
S0 maximum possible sunshine duration (h)
Ta average temperature (°C)
Tmax maximum temperature (°C)
Tmin minimum temperature (°C)
ΔT temperature diurnal(°C)
Xi the attribute variables of data set i
Yi,m measured value
Yi,c calculated value
establishment of the sunshine-based general models, only Angstrom-
Prescott model has been analyzed. Mecibah [26], proposed two general
models for Algeria based on the sunshine-based models in quadric and
cubic forms. However, the models were established based on the ob-
servation data of Algeria and could not be adopted in the case of China.
Therefore, it is necessary to establish general models of regional daily H
estimation suited for China including high accuracy to estimate the
daily H for the stations where no records of H exist.
Zoning solar radiation, comprehension of the solar radiation dis-
tribution and classification of the regions with similar solar radiation
change trends are the prerequisites for the establishment of regional
daily H estimation general models. At present, the research on the
classification of solar radiation zones focused on the selection of dif-
ferent zoning criteria [25,27–30]. As an example, based on the data of
1966–1975 from 340 SROS, Commission of the European Communities
developed the map of European solar radiation using monthly mean
daily H and S in 1984 [31]. Lau et al. [32] divided China into five zones
based on monthly mean daily S , by using data from 123 SROS in 2007.
Joseph et al. [33] proposed to take into account the outdoor tempera-
ture, radiation and air velocity as zoning criteria, and divided Mada-
gascar into three radiation zones in 2009. These zoning methods were
based on the observation data from SROS. Initially, the SROS are
classified into groups. Subsequently, the areas near to the SROS are
divided into the same groups. The above radiation zoning methodology
is based on the First Law of Geography, namely, “all attribute values on
a geographic surface are related to each other, but closer values are
more strongly related than are more distant ones” [34]. The method is
simple and easy to operate. Nevertheless, a significant limitation is that
the accuracy cannot be guaranteed.
At present, the distribution of SROS is ordinarily sparse and non-
uniform. Based on the observation data from SROS, may lead to mis-
classification problems for the areas far away from SROS, due to the
reduction of the spatial similarities of solar radiation. The problem
becomes more intense as it concerns the areas around the junctions of
two zones, resulting in high errors in the estimation of the daily H for
the stations without H . However, past research on the reduction of
these misclassifications is poor. Therefore, it is essential to propose a
method which could effectively reduce these misclassifications and
obtain accurate classification results. Nowadays, the distribution of the
general meteorological observation stations (GMOS) (since no records
169
Energy Conversion and Management 154 (2017) 168–179
Greek symbols
α Lagrange multiplier
λ latitude of stations (°)
ξ slack variables
ω normal vector
σ Gaussian noise level of standard deviation
φ (x ) attribute matrix of the input space vector x
Abbreviations
GA genetic algorithm
GMOS general meteorological observation stations
MABE mean absolute bias error
RBF radial basis function
RMSE root mean square error
rRMSE relative root mean square error (%)
RH relative humidity (%)
SMOS surface meteorological observation stations
SROS solar radiation observation stations
SVM support vector machine
SVM-C support vector machine for classification
of H exist) is much denser than that of SROS. The misclassification
issues could be effectively improved by the introduction of GMOS in the
process of classification of radiation zones. According to Mohammadi
et al. [35] and Olatomiwa et al. [36], the daily H can be predicted by
adopting Support Vector Machine (SVM) using daily sunshine rate,
daily average temperature (Ta ), daily temperature diurnal (ΔT ). The
results proved to be satisfactory. This pronounces that the SVM could
effectively describe the correlation between H and other meteor-
ological parameters. Thus, the SVM could be an effective method,
which could introduce the GMOS in the process of classification, to
achieve the misclassification reduction.
Therefore, the main objectives of this paper contain the following
parts. Firstly, a solar radiation zoning method based on the SVM should
be proposed. This method could combine the SROS and GMOS to re-
duce the misclassifications of the areas far from SROS, especially at the
areas adjacent to the junctions of two zones. Furthermore, it should
identify different solar radiation zones by the monthly mean daily H .
Secondly, the site-specific daily H estimation models should be estab-
lished and validated by the S and H from the 98 meteorological stations
in China and the best models for each solar radiation zone should be
obtained. Finally, the general models of regional H estimation should
be developed based on the best models to estimate the daily H for
different solar radiation zones in China.
2. Methods of solar radiation zoning
Solar radiation zoning is a prerequisite for the establishment of
general models of regional daily H estimation. Based on the k-means
clustering and Support Vector Machine-Genetic Algorithm (SVM-GA), a
two-step solar radiation zoning method is proposed. To distinguish the
change trend of solar radiation in different zones, the monthly mean
daily H was selected as the zoning criteria. The procedure of two-step
method is as follows:
(a) Based on the monthly mean daily H of 98 SROS, five clusters of
SROS were obtained by k-means clustering.
(b) By using the SVM-GA, the monthly mean daily S , monthly mean
daily extraterrestrial radiation (H0 ), monthly mean daily Ta ,
monthly mean daily ΔT and monthly mean daily relative humidity
(RH) of SROS were adopted as the training data set, while the
Y. Liu et al.
corresponding data of 562 GMOS were considered as the test data
set. Thus, the classification results of GMOS were obtained corre-
sponding to the clustering results of SROS.
Thus, 660 SMOS were classified into five categories to effectively
reduce the misclassifications of the distant areas from the SROS. In
particular the areas close to the junctions of two solar radiation zones
were classified and the accuracy of daily H estimation was improved,
for the stations where there is no H . In this paper, the data quantities of
k-means cluster were 1176 (98 × 12), while the data quantities of
SVM-GA used for training was 5880 (98 × 60), and the data quantities
used for testing was 33720 (562 × 60).
2.1. Clustering techniques
Clustering is also called non-supervised learning machine, since it
aims to find out the internal structure of unlabeled data set for classi-
fication. It separates the data sets into groups in a way that one sample
is much more similar to the other samples of its cluster than to samples
of other clusters [37]. Among the most popular clustering methods, k-
means clustering is one of the most used methods, commonly adopted
in statistics, pattern recognition, information retrieval and data mining
[38,39]. k-means clustering consists of the selection of k central posi-
tions in a certain data set, in such a way that the distance between each
point in the data center to their nearest central location is minimum. In
this paper, the square of Euclidean distance was adopted to measure the
distance between the points. The formula is as follows [40]:
2 1/2
⎛ d ⎢ ⎥⎞
d (Xi ,Xj ) = ⎜ ∑ ⎢x i,k −x j,k ⎥ ⎟ = ||Xi −Xj ||2
⎝ k=1 ⎣ ⎦⎠
where Xi and Xj correspond to the attribute variables of data set i and j ,
respectively.
2.2. The hybrid SVM-GA model for classification
2.2.1. Support vector machine (SVM)
Support vector machine (SVM) was proposed by Vapnik which was
developed according to statistical learning machine and structural risk
minimization [41]. Based on the limited sample information, SVM seeks
for the best trade-off between complexities of models and learning
abilities, till the best promotional capability is obtained [42,43]. It is
widely utilized in small samples, nonlinear and high dimensional pat-
tern recognition problems. Besides, it could be generalized to the pro-
blems of function fitting.
Given a data set presented by R = {x i,yi }n , where x i is the input space
vector of the sample data; x i represents the labels of classification and n
equals the number of data points. SVM for classification (SVM-C) has
the scope to construct the optimal hyperplane in feature space, by
creating the maximum classification interval. The function of SVM-C is
as follows [44]:
l mutation.
⎛1 ⎞
min ⎜ ‖ω‖2 + C ∑ ξi⎟
ω,b,ξ 2
⎝ i=1 ⎠
subject to yi (ωφ (x i )−b) ⩾ 1−ξi
where φ (x ) is the attribute matrix of the input space vector x ; ω equals
the normal vector; b stands for a scalar; l represents the number of
training data set; C > 0 , is custom penalty parameter. The greater the
value of C , the higher the restriction on misclassification samples. ξ
represents the slack variables. By introducing Lagrange multiplier, the
above optimization problem could be transformed into the following
forms:
Energy Conversion and Management 154 (2017) 168–179
n n
1
min
α 2
∑ αi αj yi yj K (x i,yi )− ∑ αi
i,j = 1 i=1 (4)
The decision function of classification is:
n
⎧
f (x ) = sgn ∑ αi yi K (xi,yi ) + b⎫⎬
⎨ i=1
⎩ ⎭ (5)
where α is Lagrange multiplier; sgn(·) equals the sign function;
K (x i,yi ) = φ (x i ) φ (x j ) , where the term K (x i,yi ) is defined as kernel func-
tion, which is depended on the two-inner vectors x i and x j in the feature
space φ (x i ) and φ (x j ) , respectively. SVM performs a non-linear data
mapping of the input variables to the future space, while a linear data
mapping from future space to the output is also executed. This is
achieved by classifying the nonlinear data set which could not be
classified in low dimensional space. However, the future space has a
high dimensional construction, is time consuming and has an expensive
computation [45]. Accordingly, it is necessary to build a non-linear
learning machine to compute the inner structure of the input variables
in the future space. Thus, a kernel function is introduced to settle this
problem, which could be effective and efficient to complete this pro-
cess.
There are four basic kernel functions in SVM: linear, polynomial,
sigmoid and radial basis function. Among them, radial basis function
(RBF) has been regarded as the best kernel function, due to its com-
putational effectiveness, reliability, simplicity, ease of adaption for
optimization, as well as its adaptability in handling factors which are
more complicated [35,36,46]. The function of RBF is as follows:
K (x i,x j ) = exp(−γ||x i−x j ||2 ) (6)
(1) 1
where γ = and σ is the Gaussian noise level of standard deviation.
− 2
2σ
The two parameters associated to the accuracies of SVM-C using RBF
kernel functions are C and γ . In this paper, the optimal values of the
two parameters are established by genetic algorithm.
2.2.2. Genetic algorithm (GA)
Genetic algorithm (GA) corresponds to a kind of stochastic search
technique based on optimization theories of survival of the fittest,
nature selection and species genetics [47]. It is based on the simulation
of the behavior of genetic variation and survival competition, to
achieve an improved problem solving and to obtain the satisfactory
solution [48]. The algorithm has implicit parallelism characteristics and
strong global searching capabilities, so that global optimization is ob-
tained in a short time. The process of genetic algorithm consists of the
following steps [49]:
(1) Select the appropriate coding mechanism to abstract the object into
a string of specific symbols in a certain order.
(2) Choose a randomly generated population.
(3) Calculate the fitness of each chromosome in the population.
(4) Create the offspring by genetic operators: selection, crossover and
(5) Judge the stopping criteria. If the stopping criteria are satisfied, the
(2)
genetic algorithm should stop. Otherwise, steps 3–4 are repeated by
using the generated offspring as the new starting population.
(3)
Fig. 1 illustrates the flow chart to obtain the optimal SVM para-
meters. In this paper, the coding mechanism of C and γ are binary
encoding; the ending evolution algebra is 30; the maximum evolution
algebra equals 300; the fold of cross-validation is 10; the fitness func-
tion represents the accuracy of SVM classifier. The LibSVM library from
Chih-Chuang Chang and Chih-Jen Lin [50] were used for the SVM-C
method.
170
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

Initial value of C, temperature (Tmin ). The periods of recording range from 6 to 30 years,
while the covering period is between 1981 and 2010. The data set was
divided into two sub-sets. One of the sub-sets derived from the first two
Coding C, in parameters thirds of the data sets, were used for training the models, and the re-
polulation maining data were adopted for validation purposes. Fig. 2 presents the
distribution of meteorological stations, including 98 SROS and 562
Initialize parameters of GA GMOS (total of 660 SMOS). Table 1 presents the detailed information of
SROS and the information of GMOS could be referred to supplementary
materials.
Train and update SVM model Errors and inconsistencies may occur in the radiation data, due to
(10-fold cross validation on Data set poor maintenance and inaccurate time counting procedure and reading
training data set) of the instruments. These errors may add to the inaccuracy of the model
performance. Therefore, a quality control of measured H , sunshine
hours, temperature and rest of the parameters was conducted to elim-
Calculate fitness function
inate spurious data and inaccurate measurements based on the fol-
lowing tests [51–53]:

Stop criterion is
satisfied
No
Select crossover mutation
Generate offspring
parameters population
Yes (a) Rejecting all meteorological data from a day if one or more data
Optimized values of C,
points were missing.
(b) Rejecting all meteorological data from a day if the H was greater
than the H0 .
Train SVM model (c) Rejecting all meteorological data from a day if the H was less than
0.03H0 .
(d) Rejecting all meteorological data from a day if the Tmin was higher
SVM-GA classification than the Tmax .

Fig. 1. Flow chart of the proposed GA-based parameters determination approach for SVM
classifier. 3.2. Solar radiation estimation models

A series of H estimation models has been established. Among them,

3. Materials and models
the sunshine-based models were the most widely used. The S con-
stitutes an ordinary observation data of the meteorological stations,
3.1. Data gathering and quality control
thus the data are widely available at the meteorological stations.
Moreover, the previous studies indicated that the sunshine-based
The studied data were provided by the Meteorological Data Center
models provide higher accuracy than the temperature-based models
of China Meteorological Administration, including daily H , daily S ,
[54,55]. Considering the availability of records, extensiveness of use,
daily average RH, daily Ta , daily maximum (Tmax ) and minimum
previous performance and simplicity, four sunshine-based models were

70°E 80°E 90°E 100°E 110°E 120°E 130°E 140°E Fig. 2. Distribution of the SMOS investigated in
this paper and solar radiation zones.

W E

S
40°N
40°N

Zone II
Zone III

30°N Zone I
30°N

Zone V Zone VI

20°N
20°N
0 100 200
km
Legend
General meteorological observation station
Solar radiation observation station Zone I

90°E 100°E 110°E 120°E

171
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

Table 1
Geographic and data records information of the studied stations.

Stations Latitude (°N) Longitude (°E) Altitude (m) Period Stations Latitude (°N) Longitude (°E) Altitude (m) Period

Training Validation Training Validation

Mohe 52.58 122.31 433.0 1993–2004 2005–2010 Juxian 35.35 118.50 107.4 1990–2003 2004–2010
Heihe 50.15 127.27 166.4 1981–2000 2001–2010 Gar 32.30 80.05 4278.6 1981–2000 2001–2010
Hailar 49.13 119.45 610.2 1981–2000 2001–2010 Naqu 31.29 92.04 4507.0 1981–2000 2001–2010
Fuyu 47.48 124.29 162.7 1993–2004 2005–2010 Lhasa 29.40 91.08 3648.9 1981–2000 2001–2010
Solon 46.36 121.13 499.7 1992–2004 2005–2010 Yushu 33.01 97.01 3681.2 1981–2000 2001–2010
Jiamusi 46.49 130.17 81.2 1981–2000 2001–2010 Guoluo 34.28 100.15 3719.0 1993–2004 2005–2010
Harbin 45.45 126.46 142.3 1981–2000 2001–2010 Chamdo 31.09 97.10 3306.0 1981–2000 2001–2010
Alaty 47.44 88.05 735.3 1981–2000 2001–2010 Ganzi 31.37 100.00 8393.5 1994–2005 2006–2010
Tacheng 46.44 83.00 534.9 1993–2004 2005–2010 Hongyuan 32.48 102.33 3491.6 1994–2005 2006–2010
Yining 43.57 81.20 662.5 1981–2000 2001–2010 Chengdu 30.42 103.50 539.3 2004–2008 2009–2010
Urumqi 43.47 87.39 935.0 1981–2000 2001–2010 Mianyang 31.27 104.44 522.7 1981–2000 2001–2010
Yanqi 42.05 86.34 1055.3 1993–2004 2005–2010 Lijiang 26.52 100.13 2392.4 1981–2000 2001–2010
Turpan 42.56 89.12 34.5 1981–2000 2001–2010 Panzhihua 26.35 101.43 1190.1 1992–2004 2005–2010
Aksu 41.10 80.14 1103.8 1993–2004 2005–2010 Tengchong 25.01 98.30 1654.6 1981–2000 2001–2010
Kashi 39.28 75.59 1289.4 1981–2000 2001–2010 Kunming 25.00 102.39 1886.5 1981–2000 2001–2010
Ruoqiang 39.02 88.10 887.7 1981–2000 2001–2010 Jinghong 22.00 100.47 582.0 1981–2000 2001–2010
Hetan 37.08 79.56 1375.0 1981–2000 2001–2010 Mengzi 23.23 103.23 1300.7 1981–2000 2001–2010
Hami 42.49 93.31 737.2 1981–2000 2001–2010 Zhengzhou 34.43 113.39 110.4 1981–2000 2001–2010
Ejinaqi 41.57 101.04 940.5 1992–2004 2005–2010 Jinghe 34.26 108.58 410.0 2006–2008 2009–2010
Dunhuang 40.09 94.41 1139.0 1981–2000 2001–2010 Nanyang 33.02 112.35 129.2 1990–2003 2004–2010
Jiuquan 39.46 98.29 1477.2 1993–2004 2005–2010 Ankang 32.43 109.02 290.8 1990–2003 2004–2010
Minqin 38.38 103.05 1367.5 1981–2000 2001–2010 Yichang 30.42 111.18 133.1 1981–2000 2001–2010
Gangcha 37.20 100.08 8301.5 1993–2004 2005–2010 Wuhan 30.37 114.08 23.1 1981–2000 2001–2010
Golmud 36.25 94.54 2807.6 1981–2000 2001–2010 shapingba 29.35 106.28 259.1 1987–2002 2003–2010
Xining 36.43 101.45 2295.2 1981–2000 2001–2010 Naxi 28.47 105.23 368.8 2003–2008 2009–2010
Lanzhou 36.03 103.53 1517.2 1981–1996 1997–2004 Jishou 28.19 109.44 208.4 1992–2004 2005–2010
Yuzhong 35.52 104.09 1874.4 2005–2008 2009–2010 Changsha 28.13 112.55 68.0 1987–2002 2003–2010
Erenhot 43.39 111.58 964.7 1981–2000 2001–2010 Guiyang 26.35 106.44 1223.8 1981–2000 2001–2010
Hailiutu 41.34 108.31 1288.0 1992–2004 2005–2010 Changning 26.25 112.24 116.6 1992–2004 2005–2010
Datong 40.06 113.20 1067.2 1981–2000 2001–2010 Guilin 25.19 110.18 164.4 1981–2000 2001–2010
Dongsheng 39.50 109.59 1461.9 1992–2004 2005–2010 Ganzhou 25.52 115.00 137.5 1981–2000 2001–2010
Yinchuan 38.29 106.13 1111.4 1981–2000 2001–2010 Huai'an 33.38 119.01 14.4 2001–2007 2008–2010
Taiyuan 37.47 112.33 778.3 1981–2000 2001–2010 Gushi 32.10 115.37 42.9 1981–2000 2001–2010
Guyuan 36.00 106.16 1753.0 1985–2002 2003–2010 Nanjing 32.00 118.48 7.1 1981–2000 2001–2010
Yan'an 36.36 109.30 958.5 1990–2003 2004–2010 Lvsi 32.04 121.36 5.5 1992–2004 2005–2010
Xifeng 35.44 107.38 1421.0 2005–2008 2009–2010 Hefei 31.47 117.18 27.0 1981–2000 2001–2010
Houma 35.39 111.22 433.8 1981–2000 2001–2010 Shanghai 31.24 121.27 5.5 1991–2004 2005–2010
Xilinhot 43.57 116.07 1003 1990–2003 2004–2010 Hangzhou 30.14 120.10 41.7 1981–2000 2001–2010
Tongliao 43.36 122.16 178.7 1981–2000 2001–2010 Tunxi 29.43 118.17 142.7 1992–2004 2005–2010
Changchun 43.54 125.13 236.8 1981–2000 2001–2010 Nanchang 28.36 115.55 46.9 1981–2000 2001–2010
Yanji 42.52 129.30 257.3 1981–2000 2001–2010 Hongjia 28.37 121.25 4.6 1992–2004 2005–2010
Chaoyang 41.33 120.27 169.9 1981–2000 2001–2010 Jian'ou 27.03 118.19 154.9 1992–2004 2005–2010
Shenyang 41.44 123.31 49.0 1981–2000 2001–2010 Fuzhou 26.05 119.17 84.0 1981–2000 2001–2010
Beijing 39.48 116.28 31.3 1981–2000 2001–2010 Guangzhou 23.10 113.20 41.0 1981–2000 2001–2010
Tianjin 39.05 117.04 2.5 1981–2000 2001–2010 Shantou 23.24 116.41 2.9 1981–2000 2001–2010
Laoting 39.26 118.53 10.5 1992–2004 2005–2010 Nanning 22.38 108.13 121.6 1981–2000 2001–2010
Dalian 38.54 121.38 91.5 1981–2000 2001–2010 Beihai 21.27 109.08 12.8 1993–2004 2005–2010
Fushan 37.30 121.15 32.6 1992–2004 2005–2010 Haikou 20.00 110.15 63.5 1981–2000 2001–2010
Jinan 36.36 117.03 170.3 1981–2000 2001–2010 Sanya 18.14 109.31 6.0 1992–2004 2005–2010

selected in this paper. around the world as a third order polynomial function:
Model 1: The Angstrom model (A-P model). This model, based on
2 3
the linear relationship between H and S , was first proposed by H = H0 ⎡
S ⎛S⎞ ⎛S⎞⎤
⎢a + b S0 + c S0 + d S0 ⎥
⎜ ⎟ ⎜ ⎟

Angstrom [56] and was later modified by Prescott [57]. It can be ex- ⎣ ⎝ ⎠ ⎝ ⎠⎦ (9)
pressed as:
Model 4: The Elagib model (exponential model). Elagib and Marnsel
S [60] established a new correlation based on S in an exponential form:
H = H0 ⎡a + b ⎤
⎢
⎣ S0 ⎥
⎦ (7)
b⎛ S ⎞
H = H0 ae ⎝ S0 ⎠ (10)
Model 2: The Ogelman Model (quadratic model). Ogelman et al.
[58] expressed the H by the following second order polynomial func-
tion of S :
3.3. Model evaluation
2
S ⎛S⎞⎤
H = H0 ⎡
⎢a + b S0 + c S0 ⎥
⎜ ⎟
The performances of different models investigated in this paper
⎣ ⎝ ⎠⎦ (8)
were evaluated by the four statistical indicators: coefficient of de-
Model 3: The Bahel model (cubic model). Bahel et al. [59] devel- termination (R2), mean absolute bias error (MABE), root mean square
oped a worldwide model based on S and H according to 48 stations error (RMSE) and relative root mean square error (rRMSE), as follows:

172
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

n 2 40
[∑i = 1 (Yi,c−Yi,c )(Yi,m−Yi,m)]
R2 = Average H (MJ/m2 d)
n
∑i = 1
n
(Yi,c−Yi,c )2 ∑i = 1 (Yi,m−Yi,m )2 (11) 36 Zone I 18 17.0 15.9
15 13.6 12.8
Zone II 12 10.4
32

Solar Radiation (MJ/m2·d)

1
n Zone III
MABE = ∑ (|Yi,m−Yi,c |)
n 28
i=1
n
∑i = 1 (Yi,m−Yi,c )2
RMSE =
n (13)
RMSE
rRMSE = × 100%
Yi,m
where Yi,m and Yi,c are measured and calculated value, respectively; n is
the number of observation data of each station. R2 corresponds to the
proportion of the regression square sum of the deviation square sum,
with higher values indicating a lower proportion and improved model
performance [61]. RMSE illustrates the short-term performance of the
models by term by term comparison of the actual deviation between the
calculated and measured values [62] and MABE provides information
on the long-term performance of the models. The different ranges of
rRMSE can be defined to show the models’ precision. The model ac-
curacy is considered excellent when rRMSE < 10%; good if
10% < rRMSE < 20%; fair if 20% < rRMSE < 30%; poor if rRMSE >
30% [63].
4. Results and discussion
4.1. Classification of solar radiation zones
The 98 SROS were divided into five clusters by k-means clustering.
Subsequently, by using the SVM-GA, the 562 GMOS were classified,
obtaining the classification results corresponding to the cluster results.
Table 2 presents the cluster results of SROS and the classification results
of GMOS could be referred to supplementary materials. By combining
the cluster and classification results, five solar radiation zones have
been obtained, as illustrated in Fig. 2.
According to Fig. 2, it was pronounced that there are numerous
GMOS close to the junctions of two zones. Only by adopting the solar
radiation data from SROS may remove these stations into inappropriate
zone classification, which could lead to inaccuracy in estimating the
daily H . In contrast, the two-step solar radiation zoning method could
effectively reduce the above-mentioned misclassifications as long as the
errors in estimating the daily H for these stations where no H data
exist.
The stations of Zone I are mainly located in Tibet, Qinghai, Ningxia,
Yunnan and Hainan, a total of 17 SROS and 107 GMOS. Zone II is
scattered in Xinjiang, Inner Mongolia and Gansu, containing 24 SROS
and 104 GMOS. Zone III was composed of 26 SROS and 152 GMOS,
distributed in East and Central China. Stations of Zone IV were mainly
9
Zone IV 6
3
(12) Zone V 0
I II III IV V
24
20
16
12
(14)
8
4
0
Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. --
Fig. 3. Monthly variations of mean daily H in five solar radiation zones.
spotted in eastern and southern coastal areas of China (20 SROS and
127 GMOS); Sichuan and Guiyang are the main part of Zone V and 11
SROS and 72 GMOS were used.
Fig. 3 illustrates the variations of the monthly mean daily H in five
solar radiation zones. The radiation of Zone I was always in a high level.
The highest value occurs in May (20.9 MJ/m2 d) and the annual mean
daily H was 17.0 MJ/m2 d. The maximum H of Zone II and Zone III
were observed in June, with monthly mean daily values of 23.6 MJ/
m2 d and 19.9 MJ/m2, respectively. The annual mean daily H were
15.9 MJ/m2 d and 13.6 MJ/m2 d, respectively. The highest H of Zone IV
and V are measured in July, with a monthly mean daily H of 18.0 MJ/
m2 d and 16.3 MJ/m2 d, respectively, while the annual mean daily H
for the same zones are 12.8 MJ/m2 d and 10.4 MJ/m2 d, respectively.
Fig. 4 illustrates the comparison of monthly mean daily S , monthly
mean daily Ta and monthly mean daily RH, between training data set
(SROS) and test data set (GMOS). Satisfying agreements were observed
between training data set and test data set. The S of Zone I presents
small fluctuations. The highest value was observed in April with
monthly mean daily values of 7.6 h; the annual mean daily value was
6.8 h. The hottest month was observed in July with a monthly mean
daily Ta of 16.1 °C. Zone II belongs to high-sunshine and low-humidity
region. The S within a whole year was always at a high level with an
annual mean daily value of 8.0 h and a maximum daily value of 9.5 h.
Annual mean RH was 51.2% while the minimum value was 38.6%. The
hottest month was August with a monthly mean Ta of 22.8 °C. The an-
nual mean daily S of Zone III was 6.7 h; the hottest month was August
with a monthly mean Ta of 23.4 °C. Zone IV and V belong to low-sun-
shine, high-humidity and hot areas. The annual mean daily S , Ta and RH
were 4.9 h and 3.6 h, 18.7 °C and 17.0 °C, 77.5% and 78.6%, respec-
tively. The RH demonstrated low variation in the whole year and was at

Table 2
Summary of the five SROS clusters derived in this study.

Zone Solar station Average H (MJ/m2 d) Station number

I Gangcha, Golmud, Gar, Naqu, Lhasa, Yushu, Guoluo, Chamdo, Ganzi, Hongyuan, Lijiang, Panzhihua, Tengchong, Kunming, 17.02 17
Jinghong, Mengzi, Sanya
II Solon, Alaty, Tacheng, Yining, Urumqi, Yanqi, Turpan, Aksu, Kashi, Ruoqiang, Hetan, Hami, Ejinaqi, Dunhuang, Jiuquan, 15.86 24
Minqin, Xining, Erenhot, Hailiutu, Datong, Dongsheng, Yinchuan, Guyuan, Xilinhot
III Mohe, Heihe, Hailar, Fuyu, Jiamusi, Harbin, Lanzhou, Yuzhong, Taiyuan, Yan’an, Xifeng, Houma, Tongliao, Changchun, Yanji, 13.58 26
Chaoyang, Shenyang, Tianjin, Beijin, Laoting, Dalian, Fushan, Jinan, Juxian, Zhengzhou, Jinghe
IV Nanyang, Wuhan, Ganzhou, Huai’an, Gushi, Nanjing, Lvsi, Hefei, Shanghai, Hangzhou, Tunxi, Nanchang, Hongjia, Jian’ou, 12.79 20
Fuzhou, Guangzhou, Shantou, Nanning, Beihai, Haikou
V Chengdu, Mianyang, Ankang, Yichang, Shapingba, Naxi, Jishou, Changsha, Guiyang, Changning, Guilin 10.33 11

173
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

Train Zone I Zone II Zone III Zone IV Zone V

Test Zone I Zone II Zone III Zone IV Zone V
18
S (h) Average S (h) 65 Ta (°C) Average Ta (°C)
45 20 18.7
16 12 8.0 17.0
8 6.8 30
6.7 16
9 55
6 4.9 15
14 12
6 4 3.6 0 7.8 7.6 8.3
Sunshine Duration (h)

45 8

Temperature (°C)
12 3 2 -15 4
0 0 35 -30 I 0 III
I II III IV V I II III IV V II III IV V I II IV V
10
25
8
15
6
5
4
-5
2
-15
Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. -- Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. --
Month Month
125
Average RH (%)
RH (%) 77.5 78.6
115 100 80
59.5 64.3
80 60 51.2
105 60
Relative Humidity (%)

40
40
95 20 20
0 I II III IV V 0 I III IV V
85 II

75

65

55

45

35
Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. --
Month
Fig. 4. Comparison of monthly mean daily data between training data set and test data set.

a high level. The hottest month was July with month mean Ta of 27.4 °C 25.1%, respectively.
and 26.2 °C, respectively. Fig. 5 illustrates the comparison of measured and calculated data in
five representative stations which correspond to five solar radiation
4.2. Site-dependent models zones. Good agreement between measured and calculated data has been
observed. The errors were in acceptable limits with rRMSE < 30%.
Nonlinear least square techniques were used to estimate the four Furthermore, it was demonstrated that the MABE, RMSE and rRMSE of
regression models of the 98 SROS using Matlab. The statistical indexes cubic model were always the least, presenting the best model perfor-
are presented in Table 3. mance.
Overall, the accuracy of the models investigated in this paper was
acceptable. The R2, MABE, RMSE and rRMSE varied from 0.61–0.93,
0.0364–0.1273, 0.0482–0.1453 and 7.5%–42.2%, respectively; the 4.3. General models
average values were 0.83, 0.0592, 0.0761 and 16.3%, respectively. The
best performance was observed in Zone II with an average According to the above analysis, the cubic model could be the
MABE = 0.0554, RMSE = 0.0735 and rRMSE = 12.8%, while Zone V general model of regional daily H estimation on horizontal surfaces to
introduced the worst model performance with MABE = 0.0684, accomplish wide applicability and satisfactory accuracy. Generally, to
RMSE = 0.0850 and rRMSE = 26.3%. improve the accuracy of general models, researchers tend to introduce
Exponential model presented the worst performance with highest geographical parameters, such as longitude, latitude and altitude, to
MABE, RMSE and rRMSE in each zone, compared to the other models. describe the variation of the trend of the empirical coefficients of the
The average values of R2 were 0.81, 0.79, 0.80, 0.83 and 0.77; re- models for different stations [24,25,64–66]. Accordingly, based on the
spectively; MABE were 0.0647, 0.0551, 0.0616, 0.0673 and 0.0796, site-dependent cubic models, considering the previous studies and
respectively; RMSE were 0.0823, 0.0729, 0.0786, 0.0835 and 0.0960, comparing the errors of regional daily H estimation by introducing
respectively; rRMSE equaled 14.7% 12.6% 16.3%, 20.6% and 29.8%, different geographical parameters, the latitude and altitude were
respectively. In contrast, cubic models demonstrated the best perfor- adopted in this paper and the general models of regional H estimation
mance by combining the least MABE, RMSE, rRMSE and highest R2 in were established. The form of general models is as follows:
each zone. The average values of R2 were 0.83, 0.80, 0.83, 0.89 and Model 5. Coupled sunshine- and geographical parameter-based
0.84, respectively; MABE were 0.0618, 0.0546, 0.0572, 0.0536 and models:
0.0643, respectively; RMSE were 0.0789, 0.0729, 0.0738, 0.0689 and
0.0812, respectively; rRMSE equaled 14.0%, 12.6%, 15.2%, 17.0% and

174
 Table 3
Performance of the proposed solar radiation estimation models for the investigated areas.
Y. Liu et al.

Zone Station R2 MABE × 100 RMSE × 100 rRMSE (%)

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

I Gangcha 0.91 0.91 0.91 0.90 0.91 4.05 4.04 4.04 4.21 14.12 5.36 5.34 5.33 5.56 5.45 8.5 8.5 8.5 8.8 8.7
Golmud 0.88 0.88 0.88 0.87 0.88 3.67 3.66 3.64 3.70 4.09 4.86 4.84 4.82 4.88 5.44 7.5 7.5 7.5 7.6 8.4
Gar 0.79 0.79 0.79 0.79 0.79 12.20 12.30 12.31 12.33 6.79 13.31 13.39 13.41 13.43 8.09 17.8 18.0 18.0 18.0 10.8
Naqu 0.81 0.81 0.81 0.80 0.81 12.61 12.65 12.65 12.73 8.17 14.42 14.44 14.45 14.53 10.00 22.6 22.6 22.6 22.8 15.7
Lhasa 0.84 0.84 0.84 0.84 0.84 5.19 5.19 5.19 5.27 6.39 6.21 6.21 6.21 6.35 7.47 9.3 9.3 9.3 9.5 11.2
Yushu 0.61 0.61 0.61 0.61 0.61 7.20 7.15 7.17 7.14 7.09 10.34 10.31 10.32 10.30 10.34 19.5 19.4 19.4 19.4 19.5
Guoluo 0.80 0.80 0.80 0.79 0.80 5.69 5.65 5.64 5.78 6.29 8.01 7.97 7.95 8.06 8.29 13.8 13.7 13.7 13.9 14.3
Chamdo 0.78 0.78 0.79 0.78 0.78 5.96 5.86 5.81 5.85 6.00 7.67 7.60 7.55 7.60 7.70 13.8 13.7 13.6 13.7 13.8
Ganzi 0.86 0.86 0.86 0.84 0.86 4.46 4.45 4.44 4.78 5.87 5.78 5.75 5.73 6.17 7.16 9.4 9.4 9.3 10.0 11.7
Hongyuan 0.89 0.89 0.89 0.85 0.89 5.38 5.32 5.31 6.14 6.70 7.01 6.96 6.95 7.96 8.19 12.2 12.1 12.0 13.8 14.2
Lijiang 0.90 0.90 0.90 0.89 0.90 4.73 4.82 4.76 4.99 5.04 6.09 6.15 6.09 6.48 6.36 11.2 11.3 11.2 11.9 11.7
Panzhihua 0.73 0.73 0.73 0.73 0.73 7.44 7.49 7.42 7.68 7.90 9.87 9.93 9.83 10.00 10.76 19.7 19.8 19.6 19.9 21.5
Tengchong 0.88 0.87 0.88 0.88 0.88 5.44 5.55 5.47 5.56 5.20 7.15 7.26 7.16 7.21 6.84 14.8 15.0 14.8 14.9 14.1
Kunming 0.89 0.90 0.90 0.86 0.90 4.87 4.76 4.63 5.66 4.77 6.45 6.34 6.17 7.38 6.32 13.4 13.2 12.8 15.4 13.2
Jinghong 0.81 0.82 0.82 0.78 0.82 4.88 4.75 4.75 5.37 4.92 6.26 6.10 6.08 6.81 6.28 13.0 12.7 12.7 14.2 13.1
Mengzi 0.79 0.79 0.80 0.76 0.80 7.02 6.92 6.79 7.54 6.58 9.68 9.60 9.51 10.21 9.18 19.5 19.4 19.2 20.6 18.5
Sanya 0.87 0.87 0.87 0.86 0.87 5.03 5.15 5.11 5.24 4.68 6.54 6.62 6.58 6.93 6.02 13.9 14.1 14.0 14.7 12.8
II Solon 0.85 0.85 0.85 0.84 0.85 4.63 4.64 4.64 4.86 4.73 6.22 6.22 6.22 6.49 6.26 10.4 10.4 10.4 10.9 10.5
Alaty 0.80 0.80 0.80 0.79 0.80 5.67 5.69 5.64 5.96 5.71 7.72 7.72 7.69 7.95 7.72 12.9 12.9 12.9 13.3 12.9
Tacheng 0.68 0.68 0.68 0.66 0.68 8.61 8.65 8.65 8.96 8.86 11.75 11.80 11.80 12.14 12.23 19.2 19.3 19.2 19.8 20.0
Yining 0.85 0.85 0.85 0.83 0.85 5.02 5.03 5.04 5.47 5.57 6.62 6.62 6.62 7.11 7.36 12.1 12.1 12.1 13.0 13.4
Urumqi 0.81 0.81 0.81 0.80 0.81 5.61 5.60 5.60 5.82 6.02 7.39 7.39 7.41 7.65 7.90 14.1 14.1 14.1 14.6 15.1
Yanqi 0.74 0.74 0.74 0.75 0.75 6.08 6.09 6.05 6.01 6.67 8.01 8.02 7.97 7.90 8.53 14.3 14.3 14.2 14.1 15.2

175
Turpan 0.80 0.81 0.81 0.81 0.81 4.59 4.48 4.47 4.45 5.06 6.12 5.96 5.97 5.90 6.54 11.5 11.2 11.2 11.1 12.3
Aksu 0.72 0.72 0.72 0.72 0.72 5.84 5.82 5.81 5.85 6.35 7.46 7.40 7.39 7.41 8.06 14.2 14.1 14.1 14.1 15.4
Kashi 0.80 0.81 0.81 0.80 0.81 6.24 6.21 6.26 6.28 6.69 7.91 7.93 7.98 8.08 8.51 15.2 15.3 15.4 15.6 16.4
Ruoqiang 0.80 0.80 0.80 0.80 0.80 4.48 4.47 4.47 4.53 4.45 6.33 6.32 6.31 6.38 6.28 11.3 11.3 11.2 11.4 11.2
Hetan 0.78 0.78 0.78 0.78 0.78 4.74 4.70 4.68 4.74 4.71 6.96 6.96 6.95 7.05 6.87 12.9 12.9 12.8 13.0 12.7
Hami 0.66 0.66 0.66 0.66 0.66 7.11 7.19 7.23 7.34 6.49 8.91 8.99 9.03 9.15 8.33 14.5 14.7 14.7 14.9 13.6
Ejinaqi 0.84 0.84 0.84 0.84 0.84 3.81 3.79 3.81 3.85 4.22 5.21 5.17 5.18 5.24 5.56 8.0 8.0 8.0 8.1 8.6
Dunhuang 0.76 0.76 0.76 0.76 0.76 5.23 5.18 5.19 5.23 5.38 6.80 6.79 6.80 6.86 6.82 10.9 10.9 10.9 11.0 10.9
Jiuquan 0.85 0.85 0.85 0.85 0.85 4.59 4.59 4.60 4.73 4.60 5.98 5.96 5.96 6.11 5.92 10.1 10.1 10.1 10.3 10.0
Minqin 0.89 0.89 0.89 0.88 0.89 4.03 4.02 4.00 4.15 4.00 5.36 5.35 5.33 5.54 5.29 8.9 8.9 8.8 9.2 8.8
Xining 0.88 0.89 0.89 0.86 0.89 5.75 5.75 5.71 6.12 5.01 7.08 7.07 7.03 7.50 6.34 13.0 13.0 12.9 13.7 11.6
Erenhot 0.77 0.77 0.77 0.76 0.77 5.19 5.22 5.22 5.37 6.12 6.98 6.99 6.99 7.11 7.78 10.9 10.9 10.9 11.1 12.1
Hailiutu 0.64 0.64 0.64 0.62 0.64 8.33 8.33 8.33 8.52 9.21 11.75 11.75 11.75 11.89 12.59 17.8 17.8 17.8 18.0 19.0
Datong 0.78 0.78 0.78 0.75 0.77 6.16 6.15 6.11 6.46 6.18 7.91 7.88 7.85 8.32 7.93 14.8 14.8 14.7 15.6 14.9
Dongsheng 0.88 0.88 0.88 0.87 0.88 4.69 4.66 4.61 4.75 5.03 6.27 6.25 6.21 6.51 6.88 11.1 11.0 11.0 11.5 12.1
Yinchuan 0.84 0.84 0.84 0.83 0.84 5.02 5.02 5.01 5.13 4.74 6.77 6.78 6.77 6.86 6.40 12.3 12.3 12.3 12.4 11.6
Guyuan 0.88 0.88 0.89 0.86 0.89 5.21 5.20 5.17 5.81 5.41 7.01 7.01 7.01 7.51 7.01 13.0 13.0 13.0 13.9 13.0
Xilinhot 0.85 0.85 0.85 0.85 0.85 4.83 4.83 4.82 4.89 4.50 6.21 6.22 6.22 6.36 5.94 10.5 10.5 10.5 10.7 10.0
III Mohe 0.83 0.83 0.83 0.80 0.83 5.62 5.57 5.56 5.94 5.46 7.33 7.29 7.27 7.74 7.10 13.4 13.3 13.3 14.2 13.0
Heihe 0.86 0.86 0.86 0.83 0.86 5.25 5.24 5.23 5.61 4.96 6.96 6.96 6.96 7.29 6.52 12.6 12.6 12.6 13.2 11.8
Hailar 0.70 0.70 0.70 0.69 0.70 7.21 7.14 7.13 7.54 7.80 9.24 9.21 9.20 9.51 9.90 16.5 16.4 16.4 17.0 17.7
(continued on next page)
Energy Conversion and Management 154 (2017) 168–179
 Table 3 (continued)

Zone Station R2 MABE × 100 RMSE × 100 rRMSE (%)

Y. Liu et al.

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

Fuyu 0.84 0.84 0.84 0.82 0.84 5.06 5.01 5.09 5.42 5.86 6.75 6.73 6.79 7.14 7.61 11.7 11.7 11.8 12.4 13.2
Jiamusi 0.86 0.86 0.86 0.83 0.86 5.03 4.97 4.97 5.52 4.86 6.57 6.52 6.52 7.13 6.40 13.1 13.0 13.0 14.2 12.7
Harbin 0.84 0.84 0.84 0.81 0.84 4.97 4.93 4.89 5.29 5.00 6.40 6.36 6.31 6.85 6.48 13.2 13.1 13.1 14.2 13.4
Lanzhou 0.86 0.86 0.86 0.84 0.86 5.07 5.08 5.01 5.34 5.54 6.48 6.48 6.36 6.84 6.98 13.7 13.7 13.4 14.4 14.7
Yuzhong 0.83 0.83 0.83 0.81 0.83 6.87 6.86 6.89 7.26 6.73 8.56 8.53 8.60 9.08 8.28 17.3 17.3 17.4 18.4 16.8

Zone Zone Station R2 MABE × 100 RMSE × 100 rRMSE (%)

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

I III Taiyuan 0.79 0.79 0.79 0.77 0.79 6.52 6.52 6.46 6.74 6.52 8.54 8.54 8.48 8.85 8.58 19.2 19.2 19.1 19.9 19.3
Yan’an 0.87 0.87 0.88 0.86 0.88 5.16 5.16 5.05 5.33 5.63 6.56 6.56 6.46 6.92 7.21 13.5 13.5 13.3 14.2 14.8
Xifeng 0.88 0.88 0.89 0.86 0.89 6.27 6.20 6.23 7.04 5.99 8.22 8.15 8.13 9.02 7.79 18.1 17.9 17.9 19.9 17.2
Houma 0.80 0.80 0.81 0.78 0.81 6.16 6.15 5.99 6.41 5.74 7.98 7.97 7.82 8.25 7.57 19.3 19.3 19.0 20.0 18.3
Tongliao 0.73 0.73 0.73 0.73 0.73 7.94 7.96 7.98 8.05 6.47 9.41 9.40 9.44 9.48 8.61 17.4 17.4 17.5 17.5 15.9
Changchun 0.85 0.85 0.85 0.82 0.85 5.14 5.15 5.14 5.67 5.15 7.12 7.12 7.10 7.59 7.02 13.8 13.9 13.8 14.8 13.7
Yanji 0.87 0.87 0.88 0.85 0.88 4.71 4.73 4.68 5.08 4.70 6.11 6.12 6.10 6.58 6.06 12.4 12.5 12.4 13.4 12.3
Chaoyang 0.70 0.70 0.70 0.69 0.70 6.96 6.96 6.93 7.15 6.97 9.21 9.23 9.20 9.29 9.22 18.2 18.3 18.2 18.4 18.3
Shenyang 0.82 0.82 0.83 0.79 0.83 5.83 5.80 5.76 6.21 5.57 7.64 7.62 7.60 7.99 7.33 15.6 15.6 15.5 16.3 15.0
Tianjin 0.88 0.88 0.89 0.87 0.89 4.88 4.87 4.68 5.01 4.58 6.22 6.20 6.02 6.37 5.90 12.7 12.7 12.3 13.1 12.1
Beijin 0.76 0.76 0.77 0.74 0.77 7.37 7.36 7.26 7.57 7.17 9.39 9.39 9.31 9.56 9.19 19.5 19.5 19.3 19.9 19.1
Laoting 0.86 0.86 0.87 0.84 0.87 4.89 4.88 4.74 5.18 5.37 6.35 6.33 6.20 6.76 6.80 12.7 12.7 12.4 13.5 13.6
Dalian 0.87 0.87 0.87 0.85 0.87 4.96 4.96 4.86 5.21 4.98 6.35 6.35 6.25 6.66 6.46 13.4 13.4 13.2 14.1 13.6
Fushan 0.89 0.89 0.89 0.87 0.89 4.47 4.47 4.39 4.79 4.45 5.71 5.71 5.62 6.14 5.67 11.8 11.8 11.6 12.7 11.7

176
Jinan 0.83 0.82 0.83 0.80 0.83 6.25 6.30 6.22 6.70 5.46 7.81 7.85 7.78 8.18 6.97 17.6 17.6 17.5 18.4 15.7
Juxian 0.88 0.88 0.89 0.85 0.89 4.75 4.75 4.53 5.23 5.46 6.09 5.98 5.72 6.83 6.66 13.2 13.0 12.4 14.8 14.5
Zhengzhou 0.85 0.86 0.87 0.82 0.87 5.79 5.62 5.34 6.43 6.56 7.38 7.22 6.88 8.05 8.04 16.8 16.5 15.7 18.4 18.3
II Jinghe 0.70 0.70 0.71 0.67 0.71 7.84 7.81 7.80 8.11 7.68 9.78 9.74 9.71 10.18 9.93 23.9 23.8 23.7 24.8 24.2
IV Nanyang 0.81 0.82 0.83 0.78 0.83 6.17 5.99 5.77 6.85 5.81 8.12 7.97 7.79 8.75 7.87 20.4 20.1 19.6 22.0 19.8
Wuhan 0.87 0.88 0.88 0.84 0.88 6.24 6.03 5.98 7.14 4.98 7.57 7.36 7.28 8.49 6.33 18.6 18.1 17.9 20.9 15.6
Ganzhou 0.85 0.85 0.86 0.81 0.86 5.74 5.56 5.44 6.75 6.01 7.89 7.74 7.60 8.80 8.27 21.5 21.1 20.7 24.0 22.5
Huai’an 0.88 0.88 0.88 0.85 0.88 5.72 5.61 5.70 6.49 5.86 7.02 6.85 6.87 7.97 7.10 16.0 15.6 15.7 18.2 16.2
Gushi 0.83 0.83 0.84 0.79 0.84 6.91 6.68 6.54 7.82 6.31 8.60 8.39 8.24 9.47 8.08 21.5 21.0 20.7 23.7 20.3
Nanjing 0.88 0.88 0.89 0.84 0.89 5.96 5.77 5.69 6.88 5.30 7.50 7.32 7.21 8.39 6.75 18.0 17.6 17.3 20.1 16.2
Lvsi 0.84 0.85 0.85 0.81 0.85 5.81 5.74 5.62 6.63 5.69 8.31 8.21 8.16 9.16 8.31 19.2 19.0 18.9 21.2 19.2
Hefei 0.87 0.87 0.88 0.83 0.88 6.16 5.94 5.88 7.09 5.24 7.66 7.45 7.34 8.58 6.78 18.4 17.9 17.6 20.6 16.3
Shanghai 0.87 0.89 0.90 0.83 0.90 5.84 5.30 5.20 7.06 6.18 7.28 6.70 6.54 8.57 7.50 17.5 16.1 15.7 20.6 18.0
Hangzhou 0.86 0.87 0.87 0.82 0.87 6.15 5.86 5.85 7.41 5.57 7.93 7.69 7.70 9.04 7.23 19.7 19.1 19.1 22.5 18.0
Tunxi 0.91 0.92 0.93 0.87 0.93 4.82 4.56 4.42 6.19 4.49 6.08 5.72 5.57 7.64 5.69 15.2 14.3 13.9 19.1 14.2
Nanchang 0.90 0.90 0.91 0.86 0.91 5.08 4.99 4.83 5.91 5.07 6.32 6.20 6.06 7.33 6.36 15.7 15.4 15.0 18.2 15.8
Hongjia 0.90 0.91 0.92 0.85 0.92 5.57 5.22 5.07 6.72 5.19 6.91 6.54 6.38 8.25 6.43 17.3 16.4 16.0 20.7 16.1
Jian’ou 0.89 0.90 0.91 0.85 0.91 6.36 6.08 6.01 7.27 4.93 8.15 7.76 7.68 9.44 6.44 20.2 19.2 19.0 23.4 16.0
Fuzhou 0.89 0.91 0.92 0.84 0.92 5.25 4.68 4.49 6.72 4.73 6.56 5.98 5.73 8.17 5.94 17.0 15.5 14.8 21.1 15.4
Guangzhou 0.87 0.88 0.89 0.83 0.89 4.99 4.81 4.58 5.87 4.41 6.29 6.09 5.82 7.26 5.63 17.0 16.4 15.7 19.6 15.2
Shantou 0.89 0.90 0.90 0.86 0.90 4.79 4.61 4.58 5.66 4.57 6.14 5.93 5.89 7.08 5.97 14.0 13.6 13.5 16.2 13.7
Nanning 0.88 0.90 0.90 0.83 0.90 5.16 4.73 4.54 6.38 4.58 6.54 6.00 5.80 7.94 5.84 16.9 15.5 15.0 20.6 15.1
Beihai 0.89 0.90 0.90 0.85 0.90 4.97 4.74 4.63 5.94 5.38 6.37 6.10 6.00 7.44 6.73 14.8 14.2 13.9 17.3 15.6
Haikou 0.85 0.85 0.86 0.81 0.86 6.80 6.50 6.45 7.72 5.48 8.42 8.18 8.10 9.29 7.12 19.9 19.4 19.2 22.0 16.9
V Chengdu 0.82 0.84 0.86 0.76 0.85 5.71 5.21 4.87 6.72 5.18 7.06 6.53 6.26 8.18 6.54 24.1 22.3 21.3 27.9 22.3
Mianyang 0.76 0.77 0.78 0.70 0.78 6.76 6.43 6.26 7.66 6.27 8.51 8.17 8.02 9.36 7.99 27.2 26.2 25.7 30.0 25.6
Ankang 0.89 0.89 0.89 0.86 0.89 5.53 5.54 5.50 6.25 5.82 6.91 6.89 6.86 7.77 7.23 19.5 19.4 19.3 21.9 20.4
III Yichang 0.80 0.82 0.83 0.76 0.83 7.88 7.48 7.46 8.91 6.53 9.76 9.44 9.40 10.62 8.53 27.1 26.2 26.1 29.5 23.7
Shapingba 0.79 0.80 0.80 0.73 0.80 6.54 6.14 6.04 7.50 6.20 8.15 7.87 7.80 9.00 7.84 29.4 28.4 28.2 32.5 28.3
Energy Conversion and Management 154 (2017) 168–179

Naxi 0.84 0.85 0.86 0.79 0.86 6.39 6.07 5.86 7.56 5.89 7.92 7.59 7.38 9.09 7.41 26.1 25.0 24.4 30.0 24.4
(continued on next page)
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

H S
= (a1 + b1cosλ + c1 h) + (a2 + b2cosλ + c2 h) ⎛ ⎞

23.6
21.9
24.9
39.1
17.8
⎜ ⎟

H0 ⎝ S0 ⎠
5
2 3
S S
30.6 + (a3 + b3cosλ + c3 h) ⎛ ⎞ + (a4 + b4 cosλ + c4 h) ⎛ ⎞
⎜ ⎟ ⎜ ⎟

26.8
31.7
42.2
24.2
S
⎝ ⎠
0 S
⎝ 0⎠ (15)
4

where λ is the latitude of stations, (°); h equals the altitude of stations,

(m). Nonlinear regression between the coefficients of site-dependent
24.1
22.8
26.2
39.6
18.6
cubic models and geographical parameters was carried out to develop
3

the general models. The empirical coefficients of general models are

presented in Table 4.
24.6
23.2
26.7
40.1
19.1
2

The statistical indexes of general models are presented in Table 3.

rRMSE (%)

The accuracy of each model was in acceptable limits. The ranges of R2,
26.1
23.7
28.0
40.3
20.3

MABE, RMSE and rRMSE were 0.61–0.93, 0.0400–0.1032,

1

0.0529–0.1327 and 8.4–39.1% with average values of 0.83, 0.0574,

0.0742 and 15.9%, respectively. Since the models were trained to use
13.27
7.27
7.36
8.01

6.32

data from the first two third of the data set and calibrated to use data
5

from the latter one third of the data set, it should be rationally that the
general models had better performances compared to that of site-de-
10.23
14.34
9.44
9.00

8.58

pendent models in some stations. General models demonstrated the best

4

performance in Zone II with the least error between measured and

calculated data. The average MABE, RMSE and rRMSE were 0.0565,
13.44
7.45
7.65
8.44

6.61

0.0746 and 13.0%, respectively. The highest MABE and RMSE were
3

observed in Zone I with values of 0.0592 and 0.0764, respectively,

13.62

while the highest rRMSE was measured in Zone V with a value of

7.58
7.77
8.61

6.77
RMSE × 100

24.7%.
2

Fig. 5 illustrates the comparison of calculated data from general

13.70

models and measured data in representative stations. A good agreement

8.05
7.95
9.03

7.19
1

has been observed between measured and calculated data. Moreover,

the difference of errors between general and site-dependent models was
10.32
5.85
5.96
6.48

5.16

slight. Additionally, the general models proposed in this paper were

5

compared with Liu et al. [25] general models. Authors have established
temperature- and sunshine-based models. The average rRMSE were
11.82
7.87
7.44
8.64

7.23

33.3% and 21.2%, 17.4% and 5.3% higher compared to the general
4

models proposed in this paper, respectively. Thus, this reveals that the
general models of regional daily H estimation proposed in this paper
10.35
5.88
6.23
6.85

5.46

demonstrated higher accuracy.

3

5. Conclusions
10.52
6.00
6.37
7.01

5.58
MABE × 100

The long-term measured RH, temperature (including Ta , Tmax , and

10.85

Tmin ) and S from 660 SMOS in China, as well as daily H from 98 SROS,
6.48
6.52
7.44

5.98

were collected and analyzed. Following conclusions have been drawn

1

through the analysis and research on these data:

0.86
0.89
0.85
0.69
0.91

(1) Combining the k-means clustering and SVM-GA, a novel two-step

5

solar radiation zoning method was proposed. According to that, five

solar radiation zones have been identified in China. The classifi-
0.79
0.84
0.75
0.65
0.84
4

cation results indicated that the method could combine the SROS
and GMOS in the process of classification. Thus, the misclassifica-
0.86
0.89
0.85
0.69
0.91

tion problems concerning the areas far away from SROS, and
3

especially for the areas close to the junctions of two zones, have
been effectively reduced.
0.86
0.88
0.84
0.68
0.90

(2) Four sunshine-based models were developed using H and S data

2

from 98 SROS. Through the analysis of statistical indexes, it was

demonstrated that cubic models presented the best performance in
0.84
0.88
0.82
0.68
0.89

each radiation zone with an average R2 = 0.84, MABE = 0.057,

R2

RMSE = 0.074 and rRMSE = 15.8%.

Changning
Changsha

(3) Nonlinear regression techniques between the empirical coefficient

Guiyang
Station

Jishou

Guilin

of site-specific cubic models and geographical parameters, in-

cluding latitude and altitude, were carried out to determine the
general models of regional daily H estimation based on radiation
Table 3 (continued)

zones. The results proved that the general models proposed in this
Zone

paper introduced a satisfactory accuracy and could represent the

models to estimate the H for the GMOS in China.
Zone

177
Y. Liu et al. Energy Conversion and Management 154 (2017) 168–179

Fig. 5. Comparison of monthly mean daily H between measured and calculated values at representative stations of each zone.

Table 4
Empirical coefficients of general models for each solar radiation zone.

Zone a1 a2 a3 a4 b1 b2 b3 b4 c1 c2 c3 c4

−6 −6 −5
I 0.27 −1.09 4.07 −1.68 −0.087 2.12 −5.30 2.37 5.41 × 10 −5.42 × 10 1.92 × 10 −1.91 × 10−5
II 0.19 −0.78 4.95 −3.67 0.089 1.82 −7.25 5.47 −3.31 × 10−5 1.73 × 10−5 2.04 × 10−5 −2.02 × 10−5
III 0.42 −0.35 3.05 −2.41 −0.31 1.50 −4.99 3.83 9.73 × 10−6 2.46 × 10−5 3.17 × 10−6 −1.10 × 10−5
IV 0.12 1.32 −2.41 1.69 0.026 −0.30 1.22 −0.94 8.64 × 10−5 5.29 × 10−4 3.63 × 10−4 −1.32 × 10−4
V 0.45 0.47 −2.29 2.83 −0.35 0.86 0.69 −2.03 2.10 × 10−5 −9.78 × 10−5 5.64 × 10−5 8.84 × 10−5

178
Y. Liu et al.
Acknowledgements
The project was financially supported by the National Key Research
Projects (Nos. 2016YFC0700400 and 2016YFC0207800) and the
National Natural Science Foundation of China (No. 51590911). The
data were kindly provided by the Meteorological Data Center of China
Meteorological Administration.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.enconman.2017.10.043.
References
[1] Dai Y. Outlook for energy supply and demand in China. In: green low-carbon de-
velopment in China. Springer International Publishing; 2013. p. 81–102.
[2] Han ZY, Fan Y, Jiao JL, Yan JS, Wei YM. Energy structure, marginal efficiency and
substitution rate: an empirical study of China. Energy 2007;32:935–42.
[3] Nazhad SHH, Lotfinejad MM, Danesh M, Amin R, Shamshirband S. A comparison of
the performance of some extreme learning machine empirical models for predicting
daily horizontal diffuse solar radiation in a region of southern Iran. Int J Remote
Sens 2017;38:6894–909.
[4] Luthander R, Widén J, Nilsson D, Palmb J. Photovoltaic self-consumption in
buildings: a review. Appl Energy 2015;142:80–94.
[5] Al-Mostafa ZA, Maghrabi AH, Al-Shehri SM. Sunshine-based global radiation
models: a review and case study. Energy Convers Manage 2014;84:209–16.
[6] Sun HW, Zhao N, Zeng XF, Yan D. Study of solar radiation prediction and modeling
of relationships between solar radiation and meteorological variables. Energy
Convers Manage 2015;105:880–90.
[7] Zang HX, Xu QS, Bian HH. Generation of typical solar radiation data for different
climates of China. Energy 2012;38:236–48.
[8] Dincer I. Renewable energy and sustainable development: a crucial review. Renew
Sustain Energy Rev 2000;4:157–75.
[9] Suehrcke H. On the relationship between duration of sunshine and solar radiation
on the earth’s surface: Angstrom’s equation revisited. Sol Energy 2000;68:417–25.
[10] Yohanna JK, Itodo IN, Umogbai VI. A model for determining the global solar ra-
diation for Makurdi, Nigeria. Renew Energy 2011;36:1989–92.
[11] Suehrcke H, Bowden RS, Hollands KGT. Relationship between sunshine duration
and solar radiation. Sol Energy 2013;92:160–71.
[12] Das A, Park JK, Park JH. Estimation of available global solar radiation using sun-
shine duration over South Korea. J Atmos Solar Terr Phys 2015;134:22–9.
[13] Manzano A, Martín ML, Valero F, Armentac C. A single method to estimate the daily
global solar radiation from monthly data. Atmos Res 2015;166:70–82.
[14] Almorox J, Hontoria C, Benito M. Models for obtaining daily global solar radiation
with measured air temperature data in Madrid (Spain). Appl Energy
2011;88:1703–9.
[15] Yacef R, Mellit A, Belaid S, Sen Z. New combined models for estimating daily global
solar radiation from measured air temperature in semi-arid climates: application in
Ghardaia, Algeria. Energy Convers Manage 2014;79:606–15.
[16] Dumas A, Andrisani A, Bonnici M, Graditi G, Leanza G, Madonia M, et al. A new
correlation between global solar energy radiation and daily temperature variations.
Sol Energy 2015;116:117–24.
[17] Hassan GE, Youssef ME, Mohamed ZE, Ali MA, Hanafy A. New temperature-based
models for predicting global solar radiation. Appl Energy 2016;179:437–50.
[18] Quej VH, Almorox J, Ibrakhimov M, Saito L. Empirical models for estimating daily
global solar radiation in Yucatán Peninsula, Mexico. Energy Convers Manage
2016;110:448–56.
[19] Elagib NA, Mansell MG. New approaches for estimating global solar radiation across
Sudan. Energy Convers Manage 2000;41:419–34.
[20] Chen RS, Kang ES, Yang JP, Lu SH, Zhao WZ. Validation of five global radiation
models with measured daily data in China. Energy Convers Manage
2004;45:1759–69.
[21] Maghrabi AH. Parameterization of a simple model to estimate monthly global solar
radiation based on meteorological variables, and evaluation of existing solar ra-
diation models for Tabouk, Saudi Arabia. Energy Convers Manage
2009;50:2754–60.
[22] Zhao N, Zeng XF, Han SM. Solar radiation estimation using sunshine hour and air
pollution index in China. Energy Convers Manage 2013;76:846–51.
[23] Ajayi OO, Ohijeagbon OD, Nwadialo CE, Olasope O. New model to estimate daily
global solar radiation over Nigeria. Sustain Energy Technol Assess 2014;5:28–36.
[24] Zhou J, Wu YZ, Gang Y. General formula for estimation of monthly average daily
global solar radiation in China. Energy Convers Manage 2005;46:257–68.
[25] Li MF, Tang XP, Wu W, Liu HB. General models for estimating daily global solar
radiation for different solar radiation zones in mainland China. Energy Convers
Manage 2013;70:139–48.
[26] Mecibah MS, Boukelia TE, Tahtah R, Gairaa K. Introducing the best model for es-
timation the monthly mean daily global solar radiation on a horizontal surface
(Case study: Algeria). Renew Sustain Energy Rev 2014;36:194–202.
[27] Willmott CJ, Vernon MT. Solar climates of the conterminous United States: a pre-
liminary investigation. Sol Energy 1980;24:295–303.
179
Energy Conversion and Management 154 (2017) 168–179
[28] Gopinathan KK, Soler A. Sky irradiation mapping. Energy Convers Manage
1993;34:159–68.
[29] Diabate L, Blanc P, Wald L. Solar radiation climate in Africa. Sol Energy
2004;76:733–44.
[30] Thermal design code for civil building (GB50176-93). Beijing: China Architecture
and Building Press; 1993 [in Chinese].
[31] Palz W, Greif J. European solar radiation Atlas: solar radiation on horizontal and
inclined surfaces. Springer; 1996.
[32] Lau CCS, Lam JC, Liu Y. Climate classification and passive solar design implications
in China. Energy Convers Manage 2007;48:2006–15.
[33] Rakoto-Joseph O, Garde F, David M, Adelard L, Randriamanantany ZA.
Development of climatic zones and passive solar design in Madagascar. Energy
Convers Manage 2009;50:1004–10.
[34] Tobler W. On the first law of geography: a reply. Ann Assoc Am Geogr
2004;94:304–10.
[35] Mohammadi K, Shamshirband S, Tong CW, Arif M, Petkovic D, Ch S. A new hybrid
support vector machine–wavelet transform approach for estimation of horizontal
global solar radiation. Energy Convers Manage 2015;92:162–71.
[36] Olatomiwa L, Mekhilef S, Shamshirband S, Mohammadi K, Petkovic D, Ch S. A
support vector machine–firefly algorithm-based model for global solar radiation
prediction. Sol Energy 2015;115:632–44.
[37] Han J, Pei J, Kamber M. Data mining: concepts and technique. Elsevier; 2011.
[38] Duda RO, Hart PE, Stork DG. Pattern classification. Wiley; 1973.
[39] Friedman J, Hastie T, Tibshirani R. The elements of statistical learning: data mining,
inference and prediction. Springer; 2001.
[40] Jain AK, Murty MN, Flynn PJ. Data clustering: a review. ACM Comput Surv (CSUR)
1999;31:264–323.
[41] Vapnik V, Golowich SE, Smola AJ. Support vector method for function approximation,
regression estimation, and signal processing. Adv Neural Inf Process Syst 1996;9:281–7.
[42] Joachims T. Text categorization with support vector machines: learning with many
relevant features. Springer; 1998.
[43] Huang C, Davis LS, Townshend JRG. An assessment of support vector machines for
land cover classification. Int J Remote Sens 2002;23:725–49.
[44] Yang HQ, Huang KZ, King I, Lyu MR. Localized support vector regression for time
series prediction. Neurocomputing 2009;72:2659–69.
[45] Shamshirband S, Mohammadi K, Chen HL, Samy GN, Petkovic D, Ma C. Daily global
solar radiation prediction from air temperatures using kernel extreme learning
machine: a case study for Iran. J Atmos Solar Terr Phys 2015;134:109–17.
[46] Wu KP, Wang SD. Choosing the kernel parameters for support vector machines by
the inter-cluster distance in the feature space. Pattern Recogn 2009;42:710–7.
[47] Goldberg DE. Genetic algorithms in search, optimization, and machine learning.
Addison-Wesley; 1989.
[48] Haupt RL, Haupt SE. Practical genetic algorithms. John Wiley and Sons; 2004.
[49] Liu SY, Tai HJ, Ding QS, Li DL, Xu LQ, Wei YG. A hybrid approach of support vector
regression with genetic algorithm optimization for aquaculture water quality pre-
diction. Math Comput Model 2013;58:458–65.
[50] Chang C-C, Lin C-J. LibSVM: a library for support vector machines 2001; Software
available at http://www.csie.ntu.edu.tw/cjlin/libsvm.
[51] Moradi I. Quality control of global solar radiation using sunshine duration hours.
Energy 2009;34:1–6.
[52] Tang WJ, Yang K, He J, Qin J. Quality control and estimation of global solar ra-
diation in China. Sol Energy 2010;84:466–75.
[53] Manzano A, Martin ML, Valero F, Armenta C. A single method to estimate the daily
global solar radiation from monthly data. Atmos Res 2015;166:70–82.
[54] Podesta GP, Nunez L, Villanueva CA, Skansi MA. Estimating daily solar radiation in
the Argentine Pampas. Agric For Meteorol 2004;123:41–53.
[55] Wu GF, Liu YL, Wang TJ. Methods and strategy for modeling daily global solar
radiation with measured meteorological data – a case study in Nanchang station,
China. Energy Convers Manage 2007;48:2447–52.
[56] Angstrom A. Solar and terrestrial radiation. Quart J Royal Meteorol Soc
1924;50:121–6.
[57] Prescott JA. Evaporation from a water surface in relation to solar radiation. Trans
Royal Soc Australia 1940;64:114–25.
[58] Ogelman BG, Ecevit A, Tasdemiroglu E. A new method for estimating solar radia-
tion from bright sunshine data. Sol Energy 1984;33:619–25.
[59] Bahel V, Bakhsh H, Srinivasan R. A correlation for estimation of global solar ra-
diation. Energy 1987;12:131–5.
[60] Elagib N, Mansell MG. New approaches for estimating global solar radiation across
Sudan. Energy Convers Manage 2000;21:271–87.
[61] Jamil B, Akhtar N. Comparative analysis of diffuse solar radiation models based on
sky-clearness index and sunshine period for humid-subtropical climatic region of
India: a case study. Renew Sustain Energy Rev 2017;78:329–55.
[62] Khorasanizadeh H, Mohammadi K, Jalilvand M. A statistical comparative study to
demonstrate the merit of day of the year-based models for estimation of horizontal
global solar radiation. Energy Convers Manage 2014;87:37–47.
[63] Jamieson PD, Porter JR, Wilson DR. A test of the computer simulation model
ARCWHEAT1 on wheat crops grown in New Zealand. Field Crops Res
1991;27:337–50.
[64] Gopinathan KK. A general formula for computing the coefficients of the correlation
connecting global solar radiation to sunshine duration. Sol Energy
1988;41:499–502.
[65] Chen RS, Lu SH, Kang ES, Yang JP, Ji XB. Estimating daily global radiation using
two types of revised models in China. Energy Convers Manage 2006;47:865–78.
[66] Li MF, Fan L, Liu HB, Guo PT, Wu W. A general model for estimation of daily global
solar radiation using air temperatures and site geographic parameters in Southwest
China. J Atmos Solar Terr Phys 2013;92:145–50.