Energy Conversion and Management 84 (2014) 209–216

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Sunshine-based global radiation models: A review and case study

Z.A. Al-Mostafa a,b,⇑, A.H. Maghrabi c, S.M. Al-Shehri a,b
a
National Astronomy Center, King Abdulaziz City for Science and Technology ‘‘KACST’’, P.O. Box 6086, Riyadh 11442, Saudi Arabia
b
King Abdullah Bin Abdulaziz Chair for Crescent Observations and Lunar Research ‘‘KACCOLR’’, King Abdulaziz University, P.O. Box 80203, Jeddah 21589, Saudi Arabia
c
National Center for Mathematics and Physics, King Abdulaziz City for Science and Technology ‘‘KACST’’, P.O. Box 6086, Riyadh 11442, Saudi Arabia

a r t i c l e i n f o a b s t r a c t

Article history: Many empirical correlations and models have been developed to estimate solar radiation around the
Received 8 January 2014 world. The most commonly used parameter for estimating global solar radiation is sunshine duration
Accepted 4 April 2014 because it can be easily and reliably measured, and data are widely available. In this paper, we report
Available online 4 May 2014
the performance of 52 sunshine models for the estimation of the monthly mean global solar radiation
on horizontal surfaces in Jouf, Saudi Arabia (29°470 N, 40°060 E, altitude 670 m). Some models are totally
Keywords: unsuitable for use in this region, and others vary in performance. The best models are identified.
Sunshine
Ó 2014 Elsevier Ltd. All rights reserved.
Global solar radiation
Jouf
Saudi Arabia

1. Introduction
Knowledge of solar radiation data is necessary for many
solar-energy applications, including the design and analysis of
energy-conversion devices and architectural design. Unfortunately,
solar radiation data are not easily available for many developing
countries because of the cost of measuring equipment and the
maintenance and calibration requirements. In situations for which
solar radiation data are scarce, models are commonly used to esti-
mate the data required for solar-energy applications. Therefore, it
is important to elaborate methods of estimating solar radiation
based on readily available meteorological data. Several empirical
models have been developed to estimate global solar radiation
(GSR) based on various meteorological and geographical parame-
ters, such as sunshine duration, mean ambient temperature,
maximum and minimum ambient temperature, relative humidity,
latitude, longitude, altitude and extraterrestrial radiation [1–8].
The most commonly used parameter for estimating GSR is
sunshine duration. Sunshine duration can be easily and reliably
measured, and data are widely available. Most sunshine-based
models for estimating average daily GSR on a monthly basis use
a modified Angström-type equation [9–12]. The original Angström
regression equation relates monthly average daily radiation to
clear-day radiation at the location in question and to the average
fraction of possible sunshine hours [9]:
⇑ Corresponding author at: National Astronomy Center, King Abdulaziz City for
Science and Technology ‘‘KACST’’, P.O. Box 6086, Riyadh 11442, Saudi Arabia. Fax:
+966 1 4813521.
E-mail address: zalmostafa@kacst.edu.sa (Z.A. Al-Mostafa).
 
H S
¼aþb ; ð1Þ
H S
where H is the monthly average daily global radiation, H is the
monthly average daily extraterrestrial radiation, S is the monthly
average daily hours of bright sunshine, S is the maximum possi-
ble sunshine duration, and a and b are empirical coefficients that
can be determined either experimentally or empirically using
some well-known models. Several types of regression models
(linear, quadratic, third-degree, and logarithmic) have been pro-
posed in the literature for estimating GSR based solely on the
S/S ratio.
The objective of the present study was to test the performance
of 52 sunshine-based models against the global radiation mea-
sured in Jouf, Saudi Arabia (29°470 N, 40°060 E, altitude 670 m). We
believe that Jouf has a clear sky; this may be attributable to the
man-made lake nearby, and we also consider it to be a virgin area
because it is far from any plants or factories [13] . The objective of
our previous study regarding global radiation was to determine the
most suitable model for predicting the measured GSR for the
capital city Riyadh [14].
2. Models
The various models used to estimate the monthly average daily
GSR on a horizontal surface exhibit different forms of dependence
on the S/S ratio. The models can be categorized according to this
dependence as follows; see Al-Mostafa et al. [14] and references
therein. The models are shown in Table 1.

http://dx.doi.org/10.1016/j.enconman.2014.04.021
0196-8904/Ó 2014 Elsevier Ltd. All rights reserved.
210 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216

Table 1
The relative percentage error for each model for each month during the year in Jouf.

Model Group Relative percentage error (%)

Jan Feb March April May June July Aug Sept Oct Nov Dec
4 1 1.98 3.99 5.15 0.31 5.83 1.58 5.35 10.58 5.96 8.11 5.21 9.05
7 1 0.45 2.75 3.82 1.09 7.08 0.41 6.39 11.56 7.27 9.28 6.51 10.56
10 1 2.61 3.73 5.27 0.43 6.51 0.95 6.74 12.39 6.44 9.18 5.69 9.03
12 1 19.32 20.44 21.63 17.63 12.07 18.23 12.03 7.46 12.08 9.95 12.70 9.83
13 1 3.28 1.31 0.02 5.13 11.73 3.91 11.34 16.93 11.84 14.20 11.05 15.02
19 1 49.53 46.60 48.93 46.32 39.37 43.62 36.39 31.64 40.17 36.35 40.60 40.79
21 1 7.04 9.33 10.26 5.68 0.22 7.21 1.00 3.76 0.02 1.76 0.73 3.13
22 1 2.82 0.28 1.06 3.99 9.48 1.81 8.16 13.13 9.83 11.40 9.05 13.63
23 1 4.05 4.03 1.88 7.08 15.62 7.52 16.83 23.48 15.29 19.03 14.47 17.40
25 1 4.04 4.68 6.39 1.61 5.67 1.73 6.29 12.09 5.50 8.52 4.76 7.79
27 1 5.98 6.09 8.00 3.30 4.32 2.98 5.34 11.29 4.05 7.36 3.31 6.01
30 1 15.85 16.72 18.09 13.91 7.82 14.27 7.53 2.60 7.90 5.46 8.55 5.72
31 1 6.91 9.89 10.52 5.96 1.15 8.07 2.49 1.92 0.80 0.51 1.50 2.74
32 1 16.80 16.59 14.25 20.08 29.49 20.42 30.70 38.06 29.16 33.23 28.25 31.63
33 1 11.70 13.86 10.60 16.24 27.49 18.57 30.57 38.83 26.67 32.17 25.77 27.71
34 1 11.80 13.01 10.16 15.79 26.14 17.30 28.44 36.22 25.51 30.37 24.63 27.10
35 1 14.21 13.05 15.34 11.02 2.85 9.65 0.89 5.20 3.38 0.51 4.06 2.30
39 1 2.68 5.33 6.20 1.41 4.05 3.23 3.03 7.88 4.32 6.01 3.59 7.77
40 1 5.61 4.01 2.51 7.74 14.89 6.85 14.83 20.75 14.91 17.61 14.10 17.93
41 1 17.98 16.95 14.94 20.81 29.53 20.46 30.08 37.10 29.38 32.92 28.46 32.33
42 1 2.68 5.33 6.20 1.41 0.31 3.23 3.03 7.88 4.32 6.01 3.59 7.77
44 1 5.73 6.42 8.08 3.39 3.71 3.55 4.26 9.93 3.56 6.48 2.82 5.84
45 1 3.20 3.74 5.51 0.69 6.76 0.72 7.46 13.37 6.57 9.68 5.81 8.82
48 1 4.09 3.32 5.65 0.83 7.80 0.25 9.56 16.09 7.32 11.30 6.56 8.83
49 1 1.35 1.80 3.66 1.26 8.95 1.32 9.75 15.83 8.73 11.98 7.96 10.97
50 1 5.34 4.88 7.05 2.30 5.92 1.49 7.42 13.71 5.52 9.25 4.77 7.17
51 1 2.17 3.94 5.20 0.36 5.99 1.44 5.69 11.03 6.06 8.37 5.31 9.02
52 1 5.80 6.97 8.42 3.74 2.89 4.32 3.05 8.47 2.84 5.43 2.11 5.39
2 2 7.25 11.68 11.52 7.01 4.20 10.91 8.23 5.79 3.29 3.90 3.97 1.39
5 2 2.44 3.98 5.37 0.54 6.16 1.28 6.43 12.17 6.12 8.81 5.37 8.86
8 2 0.76 2.89 4.05 0.85 7.06 0.44 6.70 12.12 7.17 9.41 6.41 10.31
15 2 5.24 0.15 0.24 5.35 7.99 0.43 2.61 4.77 9.19 7.93 8.42 14.82
26 2 2.44 3.70 5.05 0.21 6.13 1.30 5.01 9.61 6.29 8.19 5.54 9.24
28 2 5.56 5.67 7.51 2.79 4.57 2.75 4.79 10.13 4.42 7.26 3.68 6.56
29 2 7.61 7.11 9.22 4.59 3.40 3.84 4.65 10.62 3.03 6.55 2.30 4.67
37 2 59.74 62.78 62.29 60.37 59.96 62.77 61.96 60.96 59.43 60.06 59.71 56.92
47 2 2.97 3.93 5.48 0.66 6.23 1.21 6.14 11.48 6.20 8.76 5.45 8.79
1 3 7.55 11.15 11.59 7.08 2.63 9.44 3.90 0.52 2.25 0.99 2.94 1.45
3 3 7.63 11.56 11.71 7.20 3.56 10.31 6.03 2.35 2.92 2.55 3.61 1.26
6 3 1.67 3.41 4.59 0.28 6.22 1.22 4.28 8.09 6.51 7.94 5.75 9.72
11 3 1.56 3.41 4.53 0.34 6.19 1.24 4.42 8.52 6.49 7.95 5.74 9.77
16 3 5.53 0.10 0.54 5.67 8.12 0.55 2.14 3.79 9.40 7.81 8.63 15.16
20 3 3.93 4.94 6.41 1.63 4.92 2.42 4.09 8.73 5.00 7.11 4.26 7.72
24 3 24.00 26.84 26.99 23.26 19.88 25.49 18.58 11.38 19.59 17.92 20.16 16.26
36 3 12.82 13.49 14.44 10.07 6.02 12.60 13.73 15.98 5.04 6.90 5.71 1.54
38 3 40.12 37.00 35.18 42.08 51.80 41.17 54.83 65.78 51.51 56.58 50.43 55.42
9 4 4.74 3.80 6.29 1.50 7.61 0.08 10.53 17.94 6.94 11.63 6.19 8.12
14 4 3.92 0.23 0.51 4.57 8.90 1.27 6.15 10.21 9.58 10.10 8.80 14.13
17 4 2.81 2.45 0.39 5.52 13.94 5.96 15.70 22.77 13.56 17.52 12.76 15.65
18 4 3.83 0.86 0.05 5.16 10.66 2.91 9.01 13.77 11.06 12.48 10.27 14.92
43 4 2.38 5.06 5.92 1.12 4.35 2.96 3.31 8.17 4.62 6.30 3.88 8.08
46 4 1.78 0.70 0.83 4.23 11.13 3.35 10.32 15.36 11.22 13.46 10.44 14.13

 Group 1 (linear models): Group 1 models have a similar form to The following models were used in this study:
the Angström-type regression equation. However, the empirical Model 1: [12]
coefficients a and b differ depending on the results obtained for
first-order regression analysis.    2  3
H S S S
 Group 2 (second-order models): Some researchers have used a ¼ 0:6307  0:7251 þ 1:2089  0:4633 ð2Þ
H S S S
second-order polynomial equation in terms of the S/S ratio to
estimate the monthly average of the daily GSR on a horizontal Model 2: [15]
surface.    2  3
 Group 3 (third-order models): In the group 3 models, the H S S S
¼ 0:1520 þ 1:1334  1:1126 þ 0:4516 ð3Þ
monthly average of the daily GSR is parameterized as a function H S S S
of third-order dependence on the S/S ratio.
Model 3: [15]
 Group 4 (other models): This group contains all of the models
that differ from groups 1–3. This includes exponential, non-lin-    2
H S S
ear and logarithmic functions of the S/S ratio. The model of ¼ 0:1874 þ 0:8591  0:4764 ð4Þ
H S S
Gopinathan (model 46) [6], which is a function of latitude and
altitude in addition to the S/S ratio, is included in this group. Model 4: [16]
 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216 211

 
H S Model 20: [22]
¼ 0:3078 þ 0:4166 ð5Þ
H S    2  3
H S S S
Model 5: [16]
¼ 0:2408 þ 0:3625 þ 0:4597  0:3708 ð21Þ
H S S S
   2
H S S Model 21: [22]
¼ 0:3398 þ 0:2868 þ 0:1187 ð6Þ  
H S S H S
¼ 0:309 þ 0:368 ð22Þ
Model 6: [16] H S
   2  3
H S S S Model 22: [23]
¼ 0:4832  0:6161 þ 1:8932  1:0975 ð7Þ  
H S S S H S
¼ 0:367 þ 0:367 ð23Þ
Model 7: [17] H S
  Model 23: [23]
H S
¼ 0:324 þ 0:405 ð8Þ  
H S H S
¼ 0:233 þ 0:591 ð24Þ
Model 8: [17] H S
   2 Model 24: [24]
H S S
¼ 0:348 þ 0:320 þ 0:070 ð9Þ    2  3
H S S H S S S
¼ 2:4275 þ 11:946  16:745  7:9575 ð25Þ
Model 9: [18] H S S S
 
H S Model 25: [25]
¼ 0:0271 þ 0:3096 exp ð10Þ  
H S H S
¼ 0:2424 þ 0:5014 ð26Þ
Model 10: [19] H S
   2  3
H S S S Model 26: [25]
¼ 0:2854 þ 0:2591 þ 0:6171  0:4834 ð11Þ    2
H S S S H S S
¼ 0:0959 þ 0:9958  0:3922 ð27Þ
Model 11: [19] H S S
 
H S Model 27: [26]
¼ 0:2671 þ 0:4754 ð12Þ  
H S H S
¼ 0:215 þ 0:527 ð28Þ
Model 12: [20] H S
 
H S Model 28: [26]
¼ 0:23 þ 0:38 ð13Þ
H S    2
H S S
¼ 0:1 þ 0:874  0:255 ð29Þ
Model 13: [21] H S S
 
H S Model 29: [27]
¼ 0:318 þ 0:449 ð14Þ
H S    2
H S S
Model 14: [21] ¼ 0:148 þ 0:668  0:079 ð30Þ
H S S
 
H S Model 30: [28]
¼ 0:698 þ 0:2022 ln ð15Þ
H S  
H S
Model 15: [21] ¼ 0:2262 þ 0:418 ð31Þ
H S
   2
H S S Model 31: [29]
¼ 0:1541 þ 1:1714  0:705 ð16Þ  
H S S
H S
¼ 0:34 þ 0:32 ð32Þ
Model 16: [21] H S
   2  3
H S S S Model 32: [29]
¼ 0:1796 þ 0:9813  0:2958  0:2657 ð17Þ  
H S S S H S
¼ 0:27 þ 0:65 ð33Þ
Model 17: [21] H S

H Model 33: [30]

¼ 0:3396e0:8985ðS Þ
S
ð18Þ  
H H S
¼ 0:1538 þ 0:7874 ð34Þ
Model 18: [21] H S
 0:4146 Model 34: [30]
H S
¼ 0:7316 ð19Þ  
H S H S
¼ 0:1961 þ 0:7212 ð35Þ
Model 19: [22] H S
  Model 35: [31]
H S
¼ 0:3092 cosðuÞ þ 0:4931 ð20Þ  
H S H S
¼ 0:14 þ 0:57 ð36Þ
where u is the latitude of the site in degrees. H S
212 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216

Model 36: [31] Model 51: [41]

   2  3  
H S S S H S
¼ 0:81  3:34 þ 7:38  4:51 ð37Þ ¼ 0:297 þ 0:432 ð52Þ
H S S S H S
Model 37: [32] Model 52: [41]
   2
H S S  
¼ 0:225 þ 0:014 þ 0:001 ð38Þ H S
H S S ¼ 0:262 þ 0:454 ð53Þ
H S
Model 38: [33]
   2  3
H S S S
¼ 0:14 þ 2:52  3:71 þ 2:24 ð39Þ
H S S S 3. Data and validation methods

Model 39: [34] Monthly mean data for GSR on a horizontal surface (H) were
 
H S obtained from the Saudi Arabian Solar Radiation Atlas, which
¼ 0:313 þ 0:474 ð40Þ was published by the Saudi Arabian National Center for Science
H S
and Technology (now known as the King Abdulaziz City for Science
Model 40: [34] and Technology, or KACST) [42]. This atlas is based on data col-
 
H S lected over nine years (1971–1980) in the Kingdom of Saudi Arabia
¼ 0:307 þ 0:488 ð41Þ by the Ministry of Agriculture and Water. The data include sun-
H S
shine duration, the number of daylight hours, and global radiation
Model 41: [34] on horizontal surfaces.
 
H S First, the performance of each group of models was assessed.
¼ 0:309 þ 0:599 ð42Þ The calculated GSR values were evaluated in terms of mean
H S
percentage error (MPE) (Eq. (54)), mean bias error (MBE) (Eq.
Model 42: [35] (55)), and root-mean-square error (RMSE) (Eq. (56)), which are
  calculated as follows:
H S
¼ 0:335 þ 0:367 ð43Þ
H S
N  
1X Himeas  Hical
Model 43: [36] MPE ¼  100 ð54Þ
  N i¼1 Himeas
H S
¼ 0:388 cosðuÞ þ 0:367 ð44Þ
H S
1X N
Model 44: [37] MBE ¼ ðHimeas  Hical Þ ð55Þ
  N i¼1
H S
¼ 0:241 þ 0:488 ð45Þ
H S vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u 2
Model 45: [38]
u1 X N

  RMSE ¼ t ðHimeas  Hical Þ ; ð56Þ

N i¼1
H S
¼ 0:240 þ 0:513 ð46Þ
H S
where Himeas is the ith measured value, Hical is the ith predicted or
Model 46: [11]
calculated value, and N is the total number of observations.
  The relative percentage error (e) for an individual month is
H S
¼ 0:309 þ 0:539 cosðuÞ  0:0693Z þ 0:290 defined as follows:
H S
   
S S Himeas  Hical
þ 1:527  1:027 cosðuÞ þ 0:0926Z  0:359 e¼  100: ð57Þ
S S Himeas
ð47Þ
Following Ulgen and Hepbasli [43], we used the t-statistic
Model 47: [39] method to determine whether the estimate obtained using each
   2 equation differed significantly, at a particular confidence level,
H S S
¼ 0:195 þ 0:676  0:142 ð48Þ from the measured data. Stone [44] has proposed that the t-statis-
H S S tic can be calculated as follows:
Model 48: [40] vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  u" #ffi
H S u ðn  1ÞMBE2
¼ 0:18 þ 0:60 ð49Þ t-stat ¼ t : ð58Þ
H S RMSE2  MBE2
Model 49: [40]
Using published data, Stone [44] has demonstrated that the
  MBE and RMSE separately do not represent a reliable assessment
H S
¼ 0:24 þ 0:53 ð50Þ of model performance and can lead to false selection of the best
H S
model from a set of candidates. For equation estimates to be signif-
Model 50: [34] icant, the t-value calculated using Eq. (58) must be less than the
  value associated with the given confidence level in standard statis-
H S tical tables. In all the above statistical tests, the smaller the value is,
¼ 0:191 þ 0:571 ð51Þ
H S the better is the model performance [45].
 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216
4. Results and discussion
4.1. Monthly variations
Table 1 lists the relative percentage error for each model for
each month during the year. The performance of each model differs
from one month to another. Some models performed well in some
months and had higher e values in other months. There are nine
models with errors between -10 and 10% as follows: in group 1,
Models 39, 42, 44 and 52; in group 2, model 26; in group 3, Models
6, 11 and 20; and in group 4, model 43. Fig. 1 shows the monthly
performances for these nine models which have relative percent-
age errors (e values) between 10 and 10%.
There are twenty-two models with errors ranging from 15 to
10 and from 10 to 15 as follows: in group 1, models 4, 7, 10, 21,
22, 25, 27, 31, 45, 50 and 51; in group 2, Models 2, 5, 8, 15, 28, 29
and 47; in group 3, Models 1 and 3; and in group 4, Models 14 and 18.
There are nine models with errors ranging from 20 to 15 and
from 15 to 20 as follows: in group 1, Models 13, 30, 35, 48 and 49;
in group 3, Models 16 and 36; and in group 4, Models 9 and 46.
There are five models with errors ranging from 30 to 20 and
from 20 to 30 as follows: in group 1, Models 12, 23 and 40; in
group 3, model 24; and in group 4, model 17.
According to the percentage errors, models 19 (group 1), 37
(group 2), and 38 (group 3) are excluded because of the huge errors
they exhibited for all months. Models 32, 33, 34 and 41 (group 1)
are also excluded because of the huge errors they exhibited in
some months, mainly July, August, October and December.
4.2. Annual performance
Annually, if we refer to the MBEs, RMSEs and MPEs shown in
Table 2, models 19 (group 1), 37 (group 2), and 38 (group 3) should
be excluded from consideration because of their high percentage
values. In addition to the above excluded models because of high
MPE values (>5), we can also exclude models 3, 12, 13, 14, 15,
16, 22, 23, 30, 35, 40 and 49 (group 1), models 24 and 36 (group
3) and model 17 (group 4). Most models excluded here exhibited
high MBE values (>0.5). If the critical value for t is set to 5, then
all excluded models mentioned above are guaranteed to be
excluded, along with models 2 (group 2) and model 18 (group 4).
Group 1 contains the largest number of models (28 models),
approximately 54 % of the total number of models. The predictive
power of the models in this group varies from one model to
another.
Model 52 exhibited the best performance both among the mod-
els of this group and among all models considered in this study.
10
8
Relative percentage error (e in %)
4
2 20 group#3
0 39 group#1
1 3
-2
-4
-6
-8
-10
Fig. 1. Shows the monthly performances of models with relative percentage error (e in %) with ±10 %.
213
This model has the optimum values for all considered statistical
parameters. Models 44 and 42 exhibited the second-best predic-
tive power for the global solar radiation; their statistical parame-
ters are very similar to each other in value.
On the other hand, model 19 presented the worst performance,
according to its statistical parameters. Models 32, 33, and 34
demonstrated the second-poorest performance. Additionally, 12
models (excluding models 32, 33, 34, and 19) in this group exhib-
ited t-values above the critical value. This means that the rest of
the models demonstrated acceptable predictive power at various
levels of performance.
In group 2, four models exhibited t-values higher than the crit-
ical value (models 2, 8, 15, and 37). Model 37 presented the worst
performance, with extreme values of all statistical parameters
(MPE = 60 %, RMSE = 11, t-value = 12.2, and a and b values very
far from the optimum values). In contrast, model 29 presented
the best performance, followed by model 28. The rest of the models
in this group predicted the global radiation reasonably well.
Seven models of group 3 exhibited t-values higher than the crit-
ical value, and two of these seven models – 38 and 24 – presented
MPEs higher than 10%. The rest of the models exhibited MPEs
below 10%. Model 20 demonstrated the best performance in this
group, followed by model 6; these were the only two models with
t-values below the critical value and MPEs below 5%.
Based on the preference indicated by the statistical parameters,
model 43 in group 4 performed the best out of the models in this
group, followed by model 9. The rest of the models displayed t-val-
ues greater than the critical value, although some had MPE values
of less than 10%.
Fig. 2 presents the overall performance (average values) of the
models from each group for four parameters (MBE, RMSE, intercept
(a) and slope (b)). Except the slope, all parameters are in units of
kW h/m2.
All four groups exhibited t-values greater than the critical value.
They all presented good values of MPE and b, which differ little
from one another. Groups 3 and 1 displayed the best MBE and
MPE values, of which the group 3 values are superior. Groups 1
and 4 exhibited the lowest RMSEs. Most models yielded MPEs
below 10%, except models 12, 19, 24, 30, 32, 33, 34, 37, 38, 40
and 41. It is worth noting that the overall performance of any
group is affected by the performance of models within the group
that exhibit extreme parameter values, as observed above. Addi-
tionally, from the results presented in the previous section, it is
evident that the overall performance of specific models was
affected by bias in certain months.
To identify the best performance among the 52 models, after
the exclusion of certain models (see above discussions), we used
6 group#3
11 group#3
26 group#2
5 7 9 11 42 group#1
43 group#4
44 group#1
52 group#1
Month
214 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216

Table 2
Model regression results for measured and calculated data in Jouf.

Model Group Solar radiation (kW m2 month1) MPE (%) t b a R

Mean MBE RMSE
4 1 5.48 0.16 0.32 3.09 1.93 0.01 1.03 0.96
7 1 5.54 0.23 0.36 4.36 2.73 0.01 1.04 0.96
10 1 5.51 0.19 0.37 3.58 2.07 0.08 1.05 0.96
12 1 4.55 0.77 0.83 14.45 8.32 0.05 0.87 0.96
13 1 5.78 0.47 0.57 8.81 4.84 0.02 1.09 0.96
19 1 3.12 2.20 2.27 41.69 12.99 0.31 0.65 0.91
21 1 5.17 0.15 0.29 2.74 1.97 0.02 0.97 0.97
22 1 5.67 0.36 0.45 6.83 4.39 0.06 1.05 0.97
23 1 5.98 0.66 0.78 12.22 5.20 0.17 1.16 0.95
25 1 5.46 0.15 0.35 2.68 1.53 0.11 1.05 0.95
27 1 5.39 0.08 0.34 1.28 0.77 0.15 1.04 0.95
30 1 4.77 0.55 0.62 10.37 6.25 0.07 0.91 0.96
31 1 5.12 0.19 0.31 3.51 2.71 0.07 0.95 0.97
32 1 6.69 1.38 1.49 25.72 8.07 0.18 1.29 0.95
33 1 6.58 1.26 1.41 23.35 6.72 0.35 1.30 0.94
34 1 6.51 1.20 1.33 22.21 6.91 0.28 1.28 0.94
35 1 5.02 0.30 0.45 5.92 2.93 0.23 0.99 0.94
39 1 5.39 0.07 0.26 1.48 0.96 0.04 1.01 0.97
40 1 5.95 0.63 0.72 11.81 5.79 0.05 1.13 0.96
41 1 6.70 1.38 1.48 25.91 8.50 0.12 1.28 0.96
42 1 5.39 0.07 0.26 1.48 0.96 0.04 1.01 0.97
44 1 5.36 0.05 0.31 0.79 0.50 0.11 1.03 0.96
45 1 5.52 0.20 0.39 3.72 2.05 -0.12 1.06 0.95
48 1 5.57 0.25 0.46 4.48 2.19 0.22 1.09 0.94
49 1 5.63 0.32 0.47 5.83 3.01 0.13 1.08 0.95
50 1 5.47 0.16 0.39 2.73 1.45 0.19 1.07 0.95
51 1 5.49 0.17 0.33 3.20 1.96 0.03 1.04 0.96
52 1 5.32 0.01 0.29 0.08 0.07 0.07 1.01 0.96
2 2 4.96 0.36 0.43 6.36 5.20 0.27 0.88 0.98
5 2 5.50 0.18 0.36 3.36 1.96 0.07 1.05 0.96
8 2 5.54 0.23 0.37 4.33 2.62 0.02 1.05 0.96
15 2 5.58 0.27 0.32 5.57 4.95 0.36 0.98 0.98
26 2 5.48 0.16 0.31 3.11 2.01 0.00 1.03 0.97
28 2 5.40 0.08 0.32 1.43 0.87 0.10 1.03 0.96
29 2 5.34 0.02 0.34 0.24 0.23 0.18 1.04 0.95
37 2 2.08 3.23 3.35 60.58 12.24 0.16 0.36 0.98
47 2 5.49 0.17 0.35 3.23 1.92 0.06 1.04 0.96
1 3 5.05 0.26 0.36 4.80 3.66 0.09 0.93 0.97
3 3 5.00 0.32 0.39 5.68 4.61 0.17 0.91 0.98
6 3 5.48 0.16 0.29 3.16 2.18 0.06 1.02 0.97
11 3 5.48 0.16 0.30 3.22 2.20 0.05 1.02 0.97
16 3 5.58 0.27 0.32 5.62 4.94 0.39 0.98 0.98
20 3 5.41 0.10 0.29 1.88 1.20 0.02 1.02 0.96
24 3 4.21 1.11 1.17 20.86 9.82 0.04 0.80 0.95
36 3 4.76 0.55 0.63 9.86 5.95 0.35 0.83 0.97
38 3 7.91 2.59 2.73 48.49 9.79 0.21 1.53 0.95
9 4 5.57 0.25 0.49 4.39 1.99 0.29 1.10 0.94
14 4 5.64 0.32 0.39 6.41 4.85 0.18 1.03 0.98
17 4 5.90 0.58 0.72 10.75 4.59 0.19 1.15 0.95
18 4 5.73 0.41 0.49 7.91 5.10 0.08 1.06 0.97
43 4 5.40 0.09 0.27 1.77 1.15 0.04 1.01 0.97
46 4 5.74 0.42 0.52 7.94 4.51 0.02 1.08 0.96

MBE: mean bias error. RMSE: root-mean-square error. MPE: mean percentage error. t: is the t-statistics value; a and b are the regression constants between the measured and
the predicted values, R is the correlation coefficient between the measured and the predicted values.

an MPE cutoff of 10% to 10% to define the acceptable range and
required the t-values to lie below the critical value. Models outside
this MPE range and/or with t-values higher than the critical value
were excluded from consideration, leaving us with 20 models. Ele-
ven of these were from group 1 (models 4, 10, 21, 25, 27, 42, 44, 45,
50, 51, and 52), five were from group 2 (models 5, 26, 28, 29, and
47), two were from group 3 (models 6 and 20), and two were from
group 4 (models 9 and 43). By considering all statistical indicators,
the best-performing models were identified as follows: models 52
and 27 from group 1, models 29 and 28 from group 2, model 20
from group 3, and model 43 from group 4. Among these six models,
it was difficult to judge which performed the best, as most of them
presented very similar statistical parameters. However, we believe
that model 52 proved to be the most suitable model for predicting
the global solar radiation in Jouf.
The unsuccessful prediction of GSR radiation in Jouf by some of
the models does not mean that these formulas are inaccurate. The
models were proposed for sites with different atmospheric condi-
tions from those found in Jouf.
Several investigators have assessed the suitability of existing
models [46–49]. The general finding was that since most existing
models are empirical in nature, they are likely to be specific to the
atmospheric conditions under which they were developed, and thus
may either over- or underestimate measurements at other sites.
 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216
2 Group 1
0 Group 4
AVERAGE MBE RMSsq MPE
-2
-4
-6
Fig. 2. Presents the overall performance (average values) of the models from each group for four parameters (MBE, RMSE, intercept and slope). Except the slope, all
parameters are in units of kW h/m2.
5. Conclusions
The performance of 52 sunshine-based models of different
functional forms was assessed for estimating the monthly GSR on
horizontal surfaces in Jouf, Saudi Arabia. We found that some mod-
els are totally unsuitable for use in this region because of their poor
performance with respect to both annual and monthly data. Other
models exhibited reasonable performance and good statistical
parameters, according to the percentage error we chose, in some
months but poorer predictions and some non-significant statistical
parameters in other months. Among the models considered, the
model that was best suited for estimating the GSR for Jouf was
model 52 [41].
Acknowledgments
The authors thank the Saudi Presidency of Meteorological and
Environment (PME) for kindly meteorological information. The
authors also would like to thank King Abdulaziz City for Science
and Technology (KACST) for supporting this work. Finally our great
thanks to the reviewers for their valuable comments and
suggestions.
References
[1] Yacef R, Mellit A, Belaid S, S
ß en 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.
[2] Khorasanizadeh, Mohammadi K. Prediction of daily global solar radiation by
day of the year in four cities located in the sunny regions of Iran. Energy
Convers Manage 2013;76:385–92.
[3] Zhao NA, Zeng, Xi, Han S. Solar radiation estimation using sunshine hour and
air pollution index in China. Energy Conserv Manage 2013;76:846–51.
[4] Ajayi OO, Ohijeagbon OD, Nwadialo CE, Olasope Olumide. New model to
estimate daily global solar radiation over Nigeria. Sust Energy Technol Assess
2014;5:28–36.
[5] Khem N, Bhattarai PBK, Sapkota B, Kjeldstad B, Daponte P. Estimation of the
daily global solar radiation; Nepal experience. Measurement
2013;46:1807–17.
[6] Trabea AA, Shaltout MA. Correlation of global solar radiation with
meteorological parameters over Egypt. Renew Energy 2000;21:297–308.
[7] Maghrabi AH, Al-Mostafa ZA. Correlation of global radiation with some
meteorological parameters for Jeddah, Saudi Arabia. In: Proceedings of the
3rd international solar energy society Asia Pacific conference. Sidney,
Australia; 2009, paper WC0158.
[8] Maghrabi AH. Parameterization of a simple model to estimate monthly global
solar radiation based on meteorological variables, and evaluation of existing
solar radiation models for Tabouk, Saudi Arabia. Energy Convers Manage
2009;50:2754–60.
[9] Angstrom A. Solar and terrestrial radiation. Quart J Roy Met Soc
1924;50:121–5.
215
Group 2
Group 3
t-stat a b
[10] Duzen H, Aydin H. Sunshine-based estimation of global solar radiation on
horizontal surface at Lake Van region (Turkey). Energy Convers Manage
2012;58:35–46.
[11] Gopinathan KK. A general formula for computing the coefficients of the
correlations connecting global solar radiation to sunshine duration. Sol Energy
1988;41:499–502.
[12] Bakirci K. Models of solar radiation with hours of bright sunshine: a review.
Renew Sustain Energy Rev 2009;13:2580–8.
[13] Presidency of Meteorology and Environment (PME). Saudi Arabian annual
weather report. In Arabic; 1986.
[14] Al-Mostafa ZA, Maghrabi AH, Alshehri SM. Assessment of sunshine-based
global radiation models using data measured in Riyadh, Saudi Arabia. J Energy
Inst 2012;85:114–21.
[15] Tahran S, Sarı A. Model selection for global and diffuse radiation over the Central
Black Sea (CBS) region of Turkey. Energy Convers Manage 2005;46:605–13.
[16] Aras H, Balli O, Hepbasli A. Global solar radiation potential. Part 1. Model
development. Energy Sources B 2006;1:303–15.
[17] Ahmad F, Ulfat I. Empirical models for the correlation of monthly average daily
global solar radiation with hours of sunshine on a horizontal surface at
Karachi, Pakistan. Turk J Phys 2004;28:301–7.
[18] Almorox J, Hontoria C. Global solar radiation estimation using sunshine
duration in Spain. Energy Convers Manage 2004;45:1529–35.
[19] Ulgen K, Hepbasli A. Solar radiation models. Part 2: Comparison and
developing new models. Energy Sources 2004;26:521–30.
[20] Akpabio LE, Etuk SE. Relationship between global solar radiation and sunshine
duration for Onne, Nigeria. Turk J Phys 2003;27:161–7.
[21] Togrul IT, Togrul H. Global solar radiation over Turkey: comparison of
predicted and measured data. Renew Energy 2002;25:55–67.
[22] Ulgen K, Hepbasli A. Comparison of solar radiation correlations for Izmir,
Turkey. Int J Energy Res 2002;26:413–30.
[23] Chegaar M, Chibani A. Global solar radiation estimation in Algeria. Energy
Convers Manage 2001;42:967–73.
[24] Ertekin C, Yaldiz O. Comparison of some existing models for estimating global
solar radiation for Antalya (Turkey). Energy Convers Manage 2000;41:311–30.
[25] Ulgen K, Ozbalta N. Measured and estimated global radiation on horizontal
surface for Bornova, Izmir. In: Proceedings of the XIIth national thermal
science and technical congress. Izmit, Turkey; 2000, p. 113–18 [in Turkish].
[26] Said R, Mansor M, Abuain T. Estimation of global and diffuse radiation at
Tripoli. Renew Energy 1998;14:221–7.
[27] Aksoy B. Estimated monthly average global radiation for Turkey and its
comparison with observations. Renew Energy 1997;10:625–33.
[28] Tiris M, Tiris C, Ture IE. Correlations of monthly-average daily global, diffuse
and beam radiations with hours of bright sunshine in Gebze, Turkey. Energy
Convers Manage 1996;37:1417–21.
[29] Veeran PK, Kumar S. Analysis of monthly average daily global radiation and
monthly average sunshine duration at two tropical locations. Renew Energy
1993;3:935–9.
[30] Gopinathan KK, Soler A. A sunshine dependent global insolation model for
latitudes between 60°N and 70°N. Renew Energy 1992;2:401–4.
[31] Lewis G. An empirical relation for estimating global irradiation for Tennessee,
USA. Energy Convers Manage 1992;33:1097–9.
[32] Tasdemiroglu E, Sever R. An improved correlation for estimating solar
radiation from bright sunshine data for Turkey. Energy Convers Manage
1991;31:599–600.
[33] Samuel TDMA. Estimation of global radiation for Sri Lanka. Sol Energy
1991;47:333–7.
[34] Jain PC. A model for diffuse and global irradiation on horizontal surface. Sol
Energy 1990;45:301–8.
[35] Raja IA, Twidell JW. Distribution of global insolation over Pakistan. Sol Energy
1990;44:63–71.
216 Z.A. Al-Mostafa et al. / Energy Conversion and Management 84 (2014) 209–216
[36] Raja IA, Twidell JW. Diurnal variation of global insolation over five locations in
Pakistan. Sol Energy 1990;44:73–6.
[37] Luhanga PVC, Andringa J. Characteristic of solar radiation at Sebele, Gaborone,
Botswana. Sol Energy 1990;44:71–81.
[38] Jain S, Jain PC. A comparison of the Angström-type correlations and the
estimation of monthly average daily global irradiation. Sol Energy
1988;40:93–8.
[39] Ogelman H, Ecevit A, Tasdemiroglu E. A new method for estimating solar
radiation from bright sunshine data. Sol Energy 1984;33:619–25.
[40] Benson RB, Paris MV, Sherry JE, Justus CG. Estimation of daily and monthly
direct diffuse and global solar radiation from sunshine duration
measurements. Sol Energy 1984;32:523–35.
[41] Kholagi A, Ramadan MRI, Ali ZEH, Fattah YA. Global and diffuse solar
irradiance in Yemen (Y.A.R). Sol Energy 1983;31:55–62.
[42] Saudi Arabian National Center for Science and Technology (SANCST). Saudi
Arabian Solar Radiation Atlas; 1983.
[43] Ulgen K, Hepbasli A. Diffuse solar radiation estimation models for Turkey’s big
cities. Energy Convers Manage 2009;50:149–56.
[44] Stone RJ. Improved statistical procedure for the evaluation of solar radiation
estimation models. Sol Energy 1993;514:288–91.
[45] Elagib NA, Alvi SH, Mansell MG. Correlations between clearness index and
relative sunshine duration for Sudan. Renew Energy 1999;7:473–98.
[46] Ertekin C, Yaldiz O. Comparison of some existing models for estimating global
solar radiation for Antalya (Turkey). Energy Convers Manage
2000;41(4):311–30.
[47] Menges HO, Ertekin C, Sonmete MH. Evaluation of global solar radiation
models for Konya, Turkey. Energy Convers Manage 2006;47(18–19):3149–73.
[48] Robaa SM. Validation of the existing models for estimating global solar
radiation over Egypt. Energy Convers Manage 2009;50(1):184–93.
[49] Toğrul IT, Toğrul H. Estimation of global solar radiation under clear sky
radiation in Turkey. Renewable Energy 2002;25(1):55–67.