Renewable Energy 50 (2013) 710e721

Contents lists available at SciVerse ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Evaluation of different models to estimate the global solar radiation on inclined

surfaces
Colienne Demain*, Michel Journée, Cédric Bertrand
Royal Meteorological Institute of Belgium, Avenue Circulaire 3, B-1180 Brussels, Belgium

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

Article history: Global and diffuse solar radiation intensities are, in general, measured on horizontal surfaces, whereas
Received 27 February 2012 stationary solar conversion systems (both flat plate solar collector and solar photovoltaic) are mounted
Accepted 26 July 2012 on inclined surface to maximize the amount of solar radiation incident on the collector surface.
Available online 5 September 2012
Consequently, the solar radiation incident on a tilted surface has to be determined by converting solar
radiation from horizontal surface to the tilted surface of interest. This study evaluates the performance of
Keywords:
14 models transposing 10 min diffuse solar irradiation from horizontal to inclined surface. Solar radiation
Diffuse solar radiation
data from 8 months (April to November 2011) which include diverse atmospheric conditions and solar
Tilted surface
Daily integration
altitudes, measured on the roof of the radiation tower of the Royal Meteorological Institute of Belgium in
Statistical validation Uccle (Longitude 4.35
E, Latitude 50.79
N) were used for validation purposes. Individual model
Sky type condition performance is assessed by an inter-comparison between the calculated and measured solar global
Ground reflected albedo sensitivity radiation on a south-oriented surface tilted at 50.79
using statistical methods. The relative performance
of the different models under different sky conditions has been studied. Because statistical validation
procedures revealed that none of the considered model performs well under all types of sky conditions
a new model resulting from the coupling of three models acting under different sky conditions, is
developed for Belgium. The sensitivity of the proposed model to the ground reflection formulation is
assessed. Finally, the ability of the coupled model to handle hourly and daily data is discussed.
Crown Copyright Ó 2012 Published by Elsevier Ltd. All rights reserved.

1. Introduction collecting devices and architects designing buildings. Moreover,

given that photovoltaic technique has still a weak efficiency and
Through the centuries, scientists have found innovative ways to a high cost of production, it is important to investigate feasibility
harness the power of the Sun. Increasing the conversion of solar studies before embarking on solar central construction. However,
power into electricity is highly present on the political agenda in meteorological/radiometric stations usually measure global and
many countries, amid the push to find domestic energy sources that diffuse solar radiation received on horizontal surfaces. Conse-
are less polluting than fossil fuels. Solar energy technologies use the quently, data on inclined surfaces are generally not available and
Sun’s energy and light to provide heat, light, hot water, electricity, solar radiation incident on a tilted surface must be determined by
and even cooling, for houses, businesses and industries. They are converting solar radiation intensities measured on a horizontal
broadly characterized as either passive solar or active solar surface to that incident on the tilted surface of interest.
depending on the way they capture, convert and distribute solar If the measured normal incident beam is converted to beam
energy. Active solar techniques include the use of photovoltaic radiation on a tilted surface by geometrical relationship between
panels and solar thermal collectors to harness the energy. Passive the two surfaces, this is not the case for the diffuse component
solar techniques include orienting a building to the Sun, selecting because diffuse radiation comes from all points of the sky except
materials with favorable thermal mass or light dispersing proper- the sun. A relative large number of radiation models for inclined
ties and designing spaces that naturally circulate air. Quantitative surfaces have been proposed which include isotropic models (e.g.,
assessment of solar radiation incident on inclined surfaces is Liu and Jordan [1]) and anisotropic models (e.g., Willmot [2];
therefore very important to, e.g., engineers designing solar energy Bugler [3]; Hay [4]; Reindel et al. [5]; Perez et al. [6]; Gueymard,
[7]). Comparisons and modifications of these models and their
applications to specific regions in the world have also been
* Corresponding author. Tel.: þ32 2 3730650. undertaken (e.g., Noorian et al. [8]; Gueymard [9]; Padovan and Del
E-mail address: [email protected] (C. Demain). Col [10]; Kambezidis et al. [11]). The purpose of this work is to

0960-1481/$ e see front matter Crown Copyright Ó 2012 Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.renene.2012.07.031
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 711

evaluate the performance of 14 widely used models and to deter- angle, respectively. It should be noted that a is conventionally
mine which model (or models combination) can best be applied to measured clockwise from the South (see Fig. 1).
whole Belgium. Solar zenital angle qz is calculated as:

2. Solar radiation on tilted surfaces cos qz ¼ sin fsin d þ cos fcos dcos u: (5)

Note that Eq. (4) can be simplified in a number of instances.

The incident global solar irradiance on an inclined surface, Gb,
When the surface is horizontal (i.e. b ¼ 0.), Eq. (4) becomes:
can be divided into three components: (i) the beam component
from direct irradiation of the tilted surface, Bb, (ii) the diffuse cos qi ¼ cos qz ¼ cos dcos fcos u þ sin dsin f:
component, Db, and (iii) a reflected component that quantifies the
radiation reflected from the ground to the tilted surface, Rb, i.e., When the surface is tilted towards the equator (facing South in
the northern hemisphere):
Gb ¼ Bb þ Db þ Rb : (1)
cos qi ¼ cos dcosðf  bÞcos u þ sin dsinðf  bÞ:
Studies of clear skies (Perez et al. [12]) have led to a description
of the diffuse component being composed of an isotropic diffuse
component, Db,iso (i.e., uniform irradiance from the sky dome),
2.2. Diffuse component, Db
circumsolar diffuse component, Db,cs (i.e., resulting from the
forward scattering of solar radiation and concentrated in an area
In this study, 14 models (see Table 1) commonly used in solar
close to the sun) and a horizon brightening component, Db,hb, (i.e.,
energy to predict the diffuse radiation on a tilted surface are
concentrated in a band near the horizon and most pronounced in
compared. Each model develops the diffuse transposition factor
clear skies). A more complete version of Eq. (1) containing all
(i.e., the ratio of diffuse radiation on a tilted surface to that of
diffuse components is given in Eq. (2):
a horizontal, Rd) according to specific assumptions.


Gb ¼ Bb þ Db;iso þ Db;cs þ Db;hb þ Rb : (2)
2.2.1. Isotropic models
Isotropic sky models are simple models that assume a uniform
2.1. Direct component, Bb distribution of the diffuse radiation over the sky dome. Circumsolar
and horizon brightening parts are assumed to be zero. In reality,
As illustrated on Fig. 1, the calculation of the tilted direct a plane of tilt b facing the sun receives more diffuse radiation than
component, Bb, is purely geometrical: a plane of the same tilt in the opposite direction. Pseudo-isotropic
models introduce some corrections concerning sky anisotropy into
B
Bb ¼ BN cos qi ¼ cos qi ¼ Brb ; (3) the isotropic model.
cos qz
2.2.1.1. Liu-Jordan model (LJ). The isotropic model, also called the
where BN is the direct beam and B the direct horizontal solar irra-
Liu-Jordan model [1], was published in 1962,
diance. qz is the solar zenital angle, qi is the incidence angle of the
beam radiation on the tilted surface and rb ¼ max(0,cosqi/cosqz), is 1 þ cos b
the beam radiation conversion factor. The incidence angle qi can be Rd ¼ : (6)
2
derived by:

cos qi ¼ sin dsin fcos b  sin dcos fsin bcos a 2.2.1.2. Korokanis model (Ko). The isotropic sky assumption was
questioned in Hamilton and Jackson [13] where it was found that
þ cos dcos fcos bcos u þ cos dsin fsin bcos acos u
the sky’s southern part is responsible for 63% of the total intensity
þ cos dsin bsin asin u; (4) of diffuse radiation. Based on this finding, Korokanis [14] modified
the Liu-Jordan model as follows:
where d is the solar declination angle, f the location’s latitude and,
u the hour angle. In what follows, b and a denote the surface tilt 1
Rd ¼ ð2 þ cos bÞ
angle with respect to the horizontal plane and the surface azimuth 3

Table 1
Solar radiation models. (Db,iso, Db,cs, and Db,hb account for the isotropic, circumsolar,
and horizon brightening components, respectively).

Isotropic Anisotropic

Db,iso þ Db,cs Db,iso þ Db,cs þ Db,hb

Liu-Jordan (1962) / Bugler (1977)
/ Ma-Iqbal (1983)
/ Hay (1979)
/ Skartveit-Olseth (1979)
/ Reindel et al. (1990)
/ Temps-Coulson (1977)
/ Klucher (1979)
/ Perez et al. (1987)
/ Muneer (1997)
Korokanis (1986)
Badescu (2002)
Willmot (1982)
Fig. 1. Definition of angles used as coordinates for an element of sky radiation to an Gueymard (1987)
inclined plane of tilt b. See Table 8 for angle definitions.
712 C. Demain et al. / Renewable Energy 50 (2013) 710e721

  
for which a vertical plane oriented southwards covers 66.7% of the 1 1  cos b
ðbcos b  sin bÞ
total sky radiation. 1 þ cos b p 2
Rd1 ¼     þ  :
2 3 3
2.2.1.3. Badescu model (ba). Another pseudo-isotropic model was 1þ 1þ
2b 2b
proposed by Badescu [15],
with b ¼ 1.5.
1
Rd ¼ ð3 þ cosð2bÞÞ: 2.2.2.3. Bugler model (Bu). Bugler [3] modified the isotropic model
4
(Eq. (6)) by adding terms for the diffuse radiation coming from the
Sun’s disc and for the radiation from the rest of the sky that
2.2.2. Anisotropic models
depends on the angular height of the Sun over the horizon,
Besides the conventional isotropic approximation of Eq. (6) or
  
pseudo-isotropic formulation, more elaborated transposition 1 Bb 1 1 þ cos b
Rd ¼ ð1 þ cos bÞ þ 0:05 cos qi  : (8)
models have been developed to estimate the anisotropic effect and 2 D cos qz 2
calculate a refined value of Rd.

2.2.2.1. Willmot model (Wi). Willmot [2] introduces an anisotropic

reduction factor for tilted surface Cb, 2.2.2.4. Ma-Iqbal model (MI). Ma-Iqbal [16] proposed a model
dividing diffuse radiation into radiation emitted by the circumsolar
  region and radiation emitted by the rest of the sky. Cloudiness is
BN rb B
Rd ¼ þ Cb 1  N (7) determined on the basis of the atmosphere clearness index kt,
So So
 
1 þ cos b
Rd ¼ kt rb þ ð1  kt Þ
where Cb ¼ 1.0115e0.20293b0.080823b2 and b is in radians.So is 2
the solar constant (i.e., 1367 W m2).
where kt ¼ G/Gext is the usual clearness index (i.e., the global surface
2.2.2.2. Gueymard model (Gu). Gueymard [7] proposed to calculate solar irradiance, G, normalized by the corresponding extra-
the radiance of a partly cloudy sky as a weighted sum of the clear atmospheric irradiance, Gext) and rb is the beam radiation conver-
and overcast sky’s radiance: sion factor.
An alternative to the original Ma-Iqbal model (hereafter referred
  as Ma-Iqbal modified (MI’) model) considers the modified clearness
Rd ¼ 1  Ng Rd0 þ Ng Rd1
index, k0t , proposed by Perez et al. [17] rather than the standard
clearness index, kt, in the model formulation. The modified clear-
where Ng is the Gueymard’s weighting factor for cloud opacity. ness index k0t presents the advantage to be independent of the solar
Since cloudiness data are generally not available, Gueymard [7] zenith angle, qz:
based the function Ng on solar radiation data:
kt
k0t ¼ 0 1 (9)
Ng ¼ max½minðY; 1Þ; 0
B 1:4 C
1:031 exp@ A þ 0:1
where Y is a function of D/G given by: 9:4
0:9 þ
8   M
< D where M¼(singþ0.15(gþ3.885)1.253)1 is the optical air mass and
6:6667D=G  1:4167 if  0:227;
Y ¼ G g is the solar elevation angle in degrees (i.e., g ¼ 90qz).
:
1:2121D=G  0:1758 otherwise:
2.2.2.5. Hay model (Ha). In the Hay model [4], diffuse radiation
The clear sky radiation, Rd0, is given by a polynomial regression from the sky is composed of an isotropic component and a cir-
function of the solar altitude (or elevation angle), g, and b: cumsolar one. Horizon brightening is not taken into account. An

 anisotropy index, FHay, is used to quantify a portion of the diffuse
Rd0 ¼ exp a0 þ a1 cos q þ a2 cos2 q þ a3 cos3 q þ FðbÞGðgÞ radiation treated as circumsolar with the remaining portion of
diffuse radiation assumed to be isotropic, i.e.,
 
where coefficients ai are a function of the solar elevation angle g:   1 þ cos b
Rd ¼ FHay rb þ 1  FHay (10)
2
a0 ¼ 0:897  3:364g0 þ 3:96g02  1:909g03
a1 ¼ 4:448  12:962g0 þ 34:601g02  48:784g03 þ 27:511g04 where FHay ¼ B/Gext, is the Hay’s sky-clarity factor. The model is
a2 ¼ 2:77 þ 9:164g0  18:876g02 þ 23:776g03  13:014g04 reduced to the Liu-Jordan model (Eq. (6)) for FHay ¼ 0.
a3 ¼ 0:312  0:217g0  0:805g02 þ 0:318g03
2.2.2.6. Skartveit-Olseth model (SO). Solar radiation measurements
indicate that a significant part of sky diffuse radiation under over-
with g0 ¼ 0.01g.
cast sky conditions comes from the sky region around the zenith.
This effect vanishes when cloud cover disappears. Skartveit-Olseth
1  0:2249sin2 ðbÞ þ 0:1231sinð2bÞ 0:0342sinð4bÞ [18] modified the Hay model in order to account for this effect,
FðbÞ ¼ þ
1  0:2249 1  0:2249  
  1 þ cos b
Rd ¼ FHay rb þ Zcos b  Sðu; Ui Þ þ 1  FHay  Z ;
GðgÞ ¼ 0:408  0:323g0 þ 0:384g02  0:17g03 2
where Z ¼ max(0.32FHay,0) is the Skartveit-Olseth’s correction
The overcast sky radiation, Rd1, is a function of b in radians: factor. If FHay  0.15, then Z ¼ 0 and it reduces to the Hay model.
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 713

The term S(u,Ui) represents the effect of obstacles obscuring the and
horizon blocking a part of the diffuse radiation incident on the
inclined plane. In most cases, this term is neglected since the data
 
b
usually come from radiometric stations which are located in open P2 ¼ 1 þ sin3 :
2
terrain without any major natural or artificial obstacle.

2.2.2.7. Muneer model (mu). The Muneer model [19] estimates the
intensity of diffuse radiation as follows: 2.2.2.10. Klucher model (Kl). Klucher [23] found that the isotropic
model (Eq. (6)) gave good results for overcast skies but underesti-
Rd ¼ TM ð1  FM Þ þ FM rb ; mates irradiance under clear and partly overcast sky conditions,
characterized by an increased intensity near the horizon and near
where FM is a composite clearness function depending on the the circumsolar sky region. To overcome such a limitation, he
particular sky and azimuthal conditions. For shaded surfaces and proposed to refine the Temps-Coulson model (Eq. (12)) by intro-
sun-facing surfaces under overcast sky conditions FM is zero, while ducing a function fK determining the degree of cloud cover:
FMFHay for clear sky and partly cloudy sky conditions. The tilt factor,
    
1 þ cos b

TM, represents the ratio of the slope background diffuse irradiance
b
to the horizontal diffuse irradiance and is calculated from: Rd ¼ 1 þ fK cos2 qi cos3 qz 1 þ fK sin3 ;
2 2
    
1þcos b 2b b where fK ¼ 1(D/G)2. The Klucher model reduces to the Liu-Jordan
TM ¼ þ sin b  bcos b  psin2 : (11)
2 pð3þ2bÞ 2 model and the Temps-Coulson model when fK ¼ 0 (i.e., D/G ¼ 1) and
fK ¼ 1 (i.e., D/G ¼ 0), respectively.
Values of the radiation distribution index, b, depend on the sky
and azimuthal conditions and on the location. For European loca- 2.2.2.11. Perez model (Pe). Compared to the previously described
tions, Muneer recommends a fixed value of b ¼ 2.5 in case of models, the model proposed by Perez et al. [6] represents a more
shaded surfaces and sun-facing surfaces under overcast sky detailed analysis of the isotropic diffuse, circumsolar and horizon
conditions. A function of the anisotropic index, FHay, is established brightening radiation by using empirically derived coefficients.
for non-overcast sky conditions. The following relation has been According to this model,
derived for data coming from 14 locations all over the world
(Wlodarczyk [20]): a 1 þ cos b
Rd ¼ F1 þ ð1  F1 Þ þ F2 sin b; (13)
b 2
2b 2
¼ 0:04  0:82FHay  2:026FHay :
pð3 þ 2bÞ where F1 and F2 are sky brightness coefficients for the circumsolar
region and the region above the horizon line, respectively. Note that
Alternatively the following relation has been derived for
if F1 ¼ F2 ¼ 0, it reduces to the Liu-Jordan model (Eq. (6)). The
Southern Europe (Evseev [21]):
coefficients a and b take into account the angle of incidence of the
sun onto the inclined surface. More specifically, the angular loca-
2b 2
¼ 0:00263  0:712FHay  0:688FHay : tion of the circumsolar region is determined by the ratio a/b. They
pð3 þ 2bÞ
are calculated from the equations of solar geometry:

a ¼ maxð0; cosqi Þ;
2.2.2.8. Reindel model (Re). Reindel [5] added to the Hay model b ¼ maxðcos85
; singÞ:
(Eq. (10)) a module for the diffuse radiation coming from the region The brightness coefficients F1 and F2 are derived from the so-
near the horizon line. It was found that the intensity of diffuse called Perez coefficients:
radiation originating from this region decreases as sky cover
increases. The Reindel model reads as follows: F1 ¼ F11 ðεÞ þ F12 ðεÞD þ F13 ðεÞqz ;
   
  1 þ cos b b
Rd ¼ FHay rb þ 1  FHay 1 þ fR sin3 F2 ¼ F21 ðεÞ þ F22 ðεÞD þ F23 ðεÞqz ;
2 2
rffiffiffiffi
B where the Perez coefficients Fij are function of the sky clearness
where fR ¼ is a function modulating the intensity of diffuse
G parameter ε and the sky brightness parameter D. These factors are
radiation coming from the region near the horizon line. In fully defined by:
overcast conditions, fR equals zero, i.e., the model assumes an
isotropic diffuse radiation in the region near the horizon line. DþB 3
þ 1:041qz
ε ¼ D
3
2.2.2.9. Temps-Coulson model (TC). Assuming clear sky conditions, 1 þ 1:041qz
Temps-Coulson [22] modified the isotropic model (Eq. (6)) by
introducing two terms evaluating the diffuse radiation coming and
from the vicinity of the Sun’s disc (P1) and the sky radiation from
D
the region close to the horizon (P2), D¼ M
  Gext
1 þ cos b
Rd ¼ P1 P2 ; (12)
2 where qz is in radians and M is the optical air mass.
Many sets of Perez coefficient values have been determined by
with
different studies. In this paper, we applied the set of coefficients
P1 ¼ 1 þ cos2 qi sin3 qz ; from Perez [17] as given in Table 2.
714 C. Demain et al. / Renewable Energy 50 (2013) 710e721

Table 2 Table 3
Perez sky irradiance model coefficients. Quality criteria imposed on horizontal data.

3 F11 F12 F13 F21 F22 F23 Condition Criteria

1.000e1.065 0.008 0.588 0.062 0.060 0.072 0.022 If g > 2
G/Gcs  1.1
1.065e1.230 0.130 0.683 0.151 0.019 0.066 0.029 If g > 2
G/Gext < 1
1.230e1.500 0.330 0.487 0.221 0.055 0.064 0.026 Rb<Gb
1.500e1.950 0.568 0.187 0.295 0.109 0.152 0.014
1.950e2.800 0.873 0.392 0.362 0.226 0.462 0.001
2.800e4.500 1.132 1.237 0.412 0.288 0.823 0.056
4.500e6.200 1.060 1.600 0.359 0.264 1.127 0.131 3.2. Data quality control
>6.200 0.678 0.327 0.250 0.156 1.377 0.251
Global, direct and diffuse solar radiation data collected by the
CM11 and CH1 sensors were visually inspected and corrected in
2.3. Reflected component, Rb depth by a human operator fully devoted to this task. Because not
systematic human quality control is applied on the CNR1 data,
The classical approach to the modeling of the reflected radiation quality criteria defined in Table 3 have been imposed to the
assumes that the reflected rays are diffuse and the coefficients of recorded incident global and reflected solar radiation. Moreover, as
reflection of the beam and diffuse rays are identical. The evaluation small punctual albedo variations could subsist in the recorded data
of the ground reflected diffuse irradiance is thus dependent on the due to instrumental measurement issues, a second order linear
transposition factor for ground reflection, Rr: regression has been applied to measured albedo values (see Fig. 2).
Fitted albedo values (r1) are used to calculate the reflected trans-
Rb ¼ rGRr ; (14)
position factor (Eq. (14)). Note that fitted albedo values are rejected
where, r, is the foreground’s albedo and G is the global horizontal if smaller than 0.1 and higher than 0.35.
irradiance. Most studies assume that a constant radiance originates
from every point of the ground (i.e., the ground reflection process is
ideally isotropic). In this case, Rr can be simplified into: 4. Results

ð1  cos bÞ
Rr ¼ : (15) Prior to assess performance of the 14 selected models to estimate
2 the global solar radiation on inclined surfaces, we first evaluated the
Non-isotropic approaches have been proposed (i.e. Arnfield [24], relative importance of each component (i.e. direct, diffuse and re-
Temps and Coulson [22]) but the primary concern of the user for flected) in the recorded 8 months of south facing tilted plane global
the modeling of this component is the determination of the solar radiation in Uccle. The mean contribution ratio’s of direct,
appropriate albedo (Gueymard [7]). The ground albedo varies diffuse and reflected radiations to global radiation on the tilted
during the day for various reasons, including departure from plane as a function of the sky condition is given in Table 4. Note that
Lambert’s Law of isotropy and changes in ground properties (such sky conditions are based on the modified clearness index k0t (see Eq.
as soil’s water content or snow cover). Even the albedo of dry (9)) and derived from the global horizontal measurements.
ground varies during the day, with a minimum around noon. The It appears clearly from Table 4 that the largest contribution in
morning and afternoon albedo are in addition not symmetrical due south-oriented tilted global solar radiation is due to the diffuse
to azimuthal inhomogeneities in ground cover and possible partial component. In more than 80% of the time, the contribution of the
shading. Moreover, the early morning and late evening albedo’s are diffuse component to the global radiation is larger than 60% (and
often close to either 0 or 1, mostly due to artefact’s, such as shading reaches up to 80% in more than 50% of the time). Consequently, it is
or instrumental cosine error (Gueymard [9]). worth to choose a model estimating with accuracy the diffuse
component of global radiation.
3. Experimental data

3.1. Solar radiation data sets

The present analysis is based upon measurements recorded at

the radiometric tower of the Royal Meteorological Institute of
Belgium (RMI) located on the Brussels Uccle plateau (Longitude
4.35
East, Latitude 50.79
North at an elevation of 101 m above sea
level). Four data sets have been collected from April 01 to November
30, 2011: global and diffuse irradiance on the horizontal plane,
beam direct irradiance, and global irradiance on a 50.79
tilted
plane south-oriented. Measurements of the global and diffuse
irradiances on the horizontal plane and the global irradiance on the
inclined plan are performed by CM11-secondary standard pyran-
ometers. The direct-normal irradiance is measured with a first class
CH1 sun tracker pyrheliometer. For the diffuse measurements,
a shading disk is mounted in front of the pyranometer with the
same solid angle as the pyrheliometer blocking out the beam irra-
diance component. The irradiance measurements are acquired with
a time step of 5 s. All data are then integrated to bring them to
a 10 min time step. Reflected ground radiation is estimated by
measured incident and reflected solar radiation by a CNR1 net Fig. 2. Comparison between measured and fitted albedo during one common day:
radiometer. 12/06/2011.
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 715

Table 4 Globally, every model tends to overestimate the global solar radi-
Mean contribution ratio’s of direct, reflected and diffuse solar radiation to south- ation on the tilted surface. Absolute MBE values range from
oriented tilted global radiation as a function of sky conditions in Uccle for the
period 04/2011-11/2011. Sky conditions are estimated from the global horizontal
9.36 W m2 (3.10%) for the Bugler model (Bu in bold in Table 5) to
measurements. 58.91 W m2 (19.53%) for the Ma-Iqbal model (MI in italics in
Table 5). Absolute RMSE error values are comprised between
Sky conditions k0t Direct: Reflected: Diffuse: Data
Bb/Gb Rb/Gb GbBbRb/Gb [%]
30.97 W m2 (10.27%) for the Bu model and 146.16 W m2 (48.47%)
for the Gueymard model (Gu). The isotropic model seems to belong
All sky 0.0e1.0 0.28 0.16 0.56 100
Overcast 0.0e0.2 0.006 0.163 0.831 29 to models generating the smallest statistical errors while MI and Gu
Cloudy 0.2e0.4 0.04 0.16 0.80 26 models explode their errors. Using a modified clearness index in
Partly cloudy 0.4e0.6 0.198 0.154 0.648 25.4 the MI model enables to slightly reduce the statistical errors (e.g.
Partly clear 0.6e0.8 0.55 0.15 0.30 19.5 RMSE diminishes from 27.74% to 26.34% for MI and Ma-Iqbal
Clear 0.8e1.0 0.61 0.14 0.25 0.1
modified (MI’), respectively). Determination coefficients range
from 0.953 (Gu model) to 0.998 (Hay (Ha), Willmot (Wi), Klucher
The relative ability of the different models to predict global (Kl) and Reindel (Re) models). All skewness indexes have positive
radiation on a tilted surface was estimated by means of two values. Bu model minimizes skewness with 0.26 while Gu model
statistical error indexes: Mean Bias Error (MBE) and Root Mean maximizes it with 5.03. The maximal kurtosis is generated by Gu
Square Error (RMSE). model with 28.68 while the minimum value is calculated by
Temps-Coulson (TC) model with 0.79.
1X n
MBE ¼ ðe Þ
n i¼1 i 4.2. Models performance with respect to sky conditions

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Absolute and relative RMSE indexes for different sky conditions
u n

u1 X 2 stand on Table 6. Clearly, models performance is highly variable
RMSE ¼ t e with respect to the sky conditions. As an example, while the Gu
n i¼1 i
model appears to be quite efficient in clear sky conditions
(RMSE ¼ 16.49%), this model becomes totally unreliable in overcast
where ei¼(Gi,eGi,m) is the residual value; Gi,e are the estimated
situation (RMSE ¼ 80.50%). Perez model (Pe) best fits the
values and Gi,m represent the observed measures. A positive MBE
measurements in overcast situation (i.e. k0t < 0:2). Its absolute and
(resp. a negative MBE) means that the model tends to overestimate
relative RMSE errors are minimized with 10.16 W m2 (17.24%). Wi
(resp. underestimate) the observed measures. To obtain dimen-
model provides the smallest absolute RMSE over the cloudy
sionless statistical indicators we expressed MBE and RMSE as
ð0:2 < k0t < 0:4Þ and partly cloudy ð0:4 < k0t < 0:6Þ skies, respec-
fractions of mean solar global radiation during the respective time
tively. Values range from 23.35 W m2 (13.25%) to 35.89 W m2
interval as proposed by Davies et al. [25],
(8.84%). Finally, the Bu model provides best results under partly
clear ð0:6 < k0t < 0:8Þ and clear ðk0t > 0:8Þ skies conditions with an
MBE
MBE½% ¼ RMSE index of 43.28 W m2 (5.65%) and 42.29 W m2 (11.07%),
M respectively. It is worth pointing out that diverging models
RMSE
RMSE½% ¼ performance were reported by Evseev et al. [21] for a south-
M oriented surface tilted at 40
in Beer Sheva (Israel) highlighting
the need to evaluate the models performances prior to their
1X n
application in a specific region of the world. As an example, the Ma-
where M ¼ ðG Þ is the measures mean. In addition, the
n i ¼ 1 i;m Iqbal model was found to best perform under all sky, clear and
determination coefficient R2 as well as the skewness and kurtosis of partially cloudy conditions, whereas the Muneer model was found
the residual values were computed for each model. The determi- to give the best results for cloudy sky conditions in Israel.
nation coefficient is a measure of the linear dependence between Table 6 also reports the standard deviation of the model
the measured and estimated global solar data. A determination performances for each class of sky condition. The standard
coefficient close to one means a linear dependence between
Table 5
measured data and estimated values. The skewness statistic gives
Statistical validation for all sky conditions of the 14 models described in Section 2.2
an idea of the asymmetry of the distribution of the residuals. A as well as the coupled model (C) proposed in Section 4.3 against measurements of
positive (resp. a negative) skewness indicates that the tail on the the global solar radiation incoming on a 50.79
tilted surface south-oriented. Bold
right (resp. the left) side is longer than the left (resp. the right) side. values signify the best performing model. Values in italic signify the least
The kurtosis statistic is the measure of the peak height of the performing model.

distribution. The higher the peak is, the narrower the distribution Models MBE MBE RMSE RMSE R2 Skewness Kurtosis
range is. In practice, the best model would minimize absolute (W m2) (%) (W m2) (%)
skewness and maximize kurtosis statistic. The evaluation process Iso 11.85 3.93 31.57 10.46 0.997 0.57 9.90
was carried out in two steps. In a first step, error indexes were Bu 9.36 3.10 30.97 10.27 0.997 0.26 10.07
TC 45.02 14.93 63.64 21.10 0.995 1.14 0.79
computed regardless of the sky conditions. In a second step, the
Ha 22.05 7.31 38.94 12.91 0.998 1.04 3.36
model’s performance was evaluated according to the different sky Wi 16.86 5.59 35.07 11.63 0.998 1.00 4.35
types defined in Table 4. MI 58.91 19.53 83.65 27.74 0.995 1.38 7.06
MI’ 56.39 18.70 79.43 26.34 0.995 1.46 4.10
SO 18.95 6.28 37.61 12.47 0.996 1.15 3.91
4.1. All sky validation
Gu 49.59 16.44 146.16 48.47 0.953 5.03 28.68
Pe 25.77 8.54 53.44 17.72 0.997 1.88 4.49
Values of the various statistical errors indexes in all sky condi- Kl 27.32 9.06 45.15 14.97 0.998 1.40 4.13
tions are provided on Table 5. It should be noted that results of the Re 24.80 8.22 41.89 13.89 0.998 1.13 2.94
isotropic models (i.e., Liu-Jordan, Korokanis and Badescu models) Mu 21.71 7.20 41.39 13.72 0.997 1.37 3.75
C 5.80 1.92 29.31 9.72 0.9975 0.445 23.81
are given on a single row (Iso) as they bring very close results.
716 C. Demain et al. / Renewable Energy 50 (2013) 710e721

Table 6
Absolute and relative RMSE index in function of the modified clearness index for the 14 different models. The last line represents the standard deviation (S.D.) of the
performance of the 14 models within a class of sky condition.

Models 0:0 < k0t <0.2 0:2 < k0t <0.4 0:4 < k0t <0.6 0:6 < k0t <0.8 0:8 < k0t <1.0

[W m2] % [W m2] % [W m2] % [W m2] % [W m2] %

Iso 13.39 22.73 28.90 16.39 37.44 9.22 46.19 6.03 44.02 11.52
Bu 13.38 22.71 28.78 16.32 38.15 9.39 43.28 5.65 42.29 11.07
TC 34.22 58.08 70.54 40.02 71.30 17.55 81.64 10.65 95.99 25.12
Ha 13.41 22.75 28.77 16.32 40.79 10.04 67.57 8.81 89.45 23.41
Wi 10.54 17.89 23.35 13.25 35.89 8.84 63.29 8.26 84.75 22.18
MI 33.00 56.02 81.27 46.10 99.14 24.41 116.66 15.22 283.91 74.31
MI’ 32.32 54.87 78.25 44.39 93.84 23.11 110.77 14.46 253.37 66.32
SO 10.20 17.30 23.46 13.31 40.23 9.90 67.57 8.82 89.46 23.42
Gu 13.54 80.50 169.89 96.37 194.05 47.78 153.57 20.04 63.00 16.49
Pe 10.16 17.24 24.99 14.18 52.46 12.92 103.66 13.53 105.56 27.63
Kl 14.19 24.09 32.14 18.23 51.37 12.65 76.17 9.94 86.41 22.62
Re 13.73 23.31 30.67 17.40 45.41 11.18 71.85 9.38 93.80 24.55
Mu 13.39 22.73 29.45 16.70 43.88 10.80 72.36 9.44 91.35 23.91
S.D 4.84 3.24 15.23 2.63 15.63 6.32 10.35 3.10 0.69 0.10

deviation is smaller in case of extreme values of the clearness index where Rd1, Rd2 and Rd3 are the diffuse transposition factor calculated
(i.e., lower than 0.2 and upper than 0.8) meaning that models do by Perez (Eq. (13)), Willmot (Eq. (7)) and Bugler (Eq. (8)) models,
not widely differ in those conditions. In contrast, they exhibit very respectively, and w1, w2 and w3 are weighting functions defined as,
different levels of performance in the intermediate classes of k0t
(i.e., between 0.4 and 0.8) (see last line on Fig. 6). 1
w1 ¼ ;
1 þ esðkt 0:2Þ
0

4.3. Coupled model

w2 ¼ 1  w1  w3 ;
Because the validation procedure based on statistical analysis
revealed that none of the considered model performs well under all and
types of sky conditions, we propose a new model (C) for Belgium
that combines different models which best perform under specific 1
w3 ¼
1 þ esðkt 0:6Þ
0
sky conditions. Relying on conclusions drawn from Table 6, this
new coupled model is defined by combining three models:
8 where d indicates the slope of the transition between each model.
< Perez k0t < 0:2 The weights w1, w2 and w3 are actually exclusively defined by the
Willmot 0:2 < k0t < 0:6 single parameter s which is introduced to ensure a smooth tran-
:
Bugler k0t >0:6 sition between the 3 models. As illustrated on Fig. 3, s enables
to control the transition speed between the models. Values of
In order to ensure a smooth transition between models, the
s between 50 and 150 may ensure a reasonable trade-off between
diffuse transposition factor Rd,b used in the coupled model is
smoothness and rapid transition between the models. As shown in
defined as :
Fig. 4, within this range of values, the impact of s on the overall
Rd;b ¼ w1 Rd1 þ w2 Rd2 þ w3 Rd3 performance of the coupled model is limited (i.e., an RMSE varia-
tion of 1.4%). By comparison the RMSE variation between the

Fig. 3. Evolution of the weighted functions w1, w2 and w3 as a function of the modified Fig. 4. Variation of the coupled model statistical errors (i.e., RMSE in [%] as a function
clearness index for 4 different values of s (i.e., 20, 50, 100 and 200). of different values of s (i.e., 20, 35, 50, 100, 150, 200, 250, 500 and 1000).
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 717

Fig. 5. Daily time evolution of the global solar irradiance computed by the coupled model (blue curve) compared to measured data (red curve) for the 15th of (A) April, (B) May, (C)
June, (D) July, (E) August, (F) September, (G) October and (H) November. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of
this article.)
718 C. Demain et al. / Renewable Energy 50 (2013) 710e721

Fig. 6. Performance of the proposed coupled model for different sky conditions: (A) all sky, (B) overcast, (C) cloudy, (D) partly cloudy, (E) partly clear and (F) clear sky, respectively.

different models performances reported in Table 5 accounts for square linear-fit has been added to each scatter-plot. The dispersion
38.2%. In the following an intermediate value of 100 is chosen for around regression line provides some insight on the accuracy of the
the parameter s. model which can be quantified by the determination coefficient
Fig. 5 presents the daily time evolution of global irradiance on R2 (see Table 7).
a south-oriented inclined surface tilted at 50.79
for the 15th day of Globally, the C model presents the tendency to overestimate the
eight different months (April to November 2011) as measured by tilted global solar radiation predicted values in comparison to in
the pyranometer (red curve) and computed by the C model (blue situ measurements. When compared to the best individual model,
curve). As we can see, while the simulated values slightly over- Table 5 indicates that MBE indexes are sharply diminished (from
estimate measurements (see Table 5 for numeric results), variations 9.36 W m2 (3.10%) for the Bu model to 5.80 W m2 (1.92%) for the
in the recorded tilted global solar radiation are very well repro- C model. In partly cloudy and clear sky conditions data are more
duced by the C model. Fig. 6 displays individual scatter plots for the sharply merged along the regression line unlike overcast and
global solar radiation as a function of the sky condition. A least cloudy sky conditions.
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 719

Table 7 Table 9
Coefficients of determination R2 and RMSE indexes as a function of the sky condi- Sensitivity of the coupled model to the ground reflected component formulation.
tions from April to November 2011 for the coupled model.
Albedo MBE MBE RMSE RMSE
Sky k0t R2 RMSE [W m2] RMSE [%] [5%] [W m2] [%] [W m2]
All sky 0.0e1.0 0.9975 29.31 9.72 Isotropic Daily þ seasonal r1 1.92 5.80 9.72 29.31
Overcast 0.0e0.2 0.9974 12.00 21.62 Daily r2 2.25 6.79 9.25 27.89
Cloudy 0.2e0.4 0.9970 19.26 13.34 Constant r3 2.49 7.50 9.62 29.03
Partly cloudy 0.4e0.6 0.9974 27.87 9.69 Anisotropic Daily þ seasonal r1 4.47 13.48 11.11 33.51
Partly clear 0.6e0.8 0.9988 37.76 7.22 Daily r2 4.89 14.76 10.67 32.19
Clear 0.8e1.0 0.9989 44.56 8.02 Constant r3 5.13 15.47 11.11 33.51
r4 12.38 37.35 18.06 54.48

5. Sensitivity analysis
r4 ¼ a expðbqz Þ (19)
Sensitivity analysis is an important component in thorough
empirical validation. Two types of sensitivity analysis were per- where a and b are parameters depending on solar elevation angle
formed: sensitivity to (1) the ground reflected transposition factor (see Table 8).
and (2) the time integration of the input data (i.e., time step). Results of the various sensitivity experiments (3 albedo
formulations and 2 anisotropic models) are provided in terms of
MBE and RMSE in Table 9.
5.1. Ground reflection sensitivity

5.1.1. Isotropy process

Because measurements of the ground surface albedo are not
Statistical error indexes showed in Table 9 recommend the use
always available, sensitivity of the C model to the ground albedo
of the isotropic reflection model instead of anisotropic formulation.
formulation in the ground reflected transposition factor Rr has been
In fact, anisotropic reflection mostly occurs when direct component
carried out. This was done by using the C model with three different
predominates the other ones but in Belgium diffuse component is
albedo representations. A first formulation (r1) accounts for the
mostly present which generates a lower error value for isotropic
diurnal and seasonal albedo variations and corresponds to the
reflection than for anisotropy. Anisotropic MBE and RMSE results
fitted albedo value determined from measured data. The second
are higher for Arnfield et al. [27] formulation than for Temps and
formulation (r2) accounts for its diurnal variation only and relies on
Coulson [22].
the formulation proposed by Dickinson [26]:

1þF 5.1.2. Albedo formulations

r2 ¼ r0 (16) Accounting for the daily or seasonal albedo variations allow
1 þ 2Fcos qz
more accurate predicted values in global solar radiation than using
where r0 ¼ 0.23 is the albedo value at a solar zenital angle of 60
. a fixed value of the albedo. The MBE index is minimized for daily
F is a site dependent factor and equals 0.1 for urban site. and seasonal albedo variations while the RMSE is minimized in case
The last formulation (r3) considers a constant albedo value of daily albedo variations.
corresponding to the mean value of the 8 months of recorded
albedo data. 5.2. Time integration sensitivity
P
n
ri The sensitivity of the proposed model for Belgium to the input
r3 ¼ (17)
n data time integration period has been investigated. This was done
by running the C model with hourly and daily integrated input data
where n is the number of days during the 8 months period and
in addition to the previous 10 min values. Note that for these
r3 ¼ r0 for Uccle. In addition to the albedo formulation, we have
computations the hourly and daily irradiation values (Wh m2)
also investigated if considering anisotropy in the ground reflection
were obtained by integrating the 10 min values over each hour and
process could significantly modify the results of the C model. As an
each day, respectively. For hourly integration, two cases have been
example, under clear sky conditions, direct radiation is highly
considered for the angles: (1) instantaneous mid-hourly values
present. Anisotropic reflections may occur in clear sky conditions
(hereafter referred to as HOUR mid) and (2) integrated hourly
and especially when soil surface is covered with layer of water or
values (hereafter referred to as HOUR int). More details on the
when plants have glossy leaves. Two different approaches to model
angular time integration can be found in Appendix A. Table 10
the anisotropic reflection Rr have been evaluated: the Temps-
compares the coupled model performance to 10 min, hourly and
Coulson [22] anisotropic reflected factor Rr defined as:
daily input data as a function of the integrated output time horizon.
    It is worth pointing out that in Table 10 the columns indicate
ð1  cos bÞ qz
Rr ¼ 1 þ sin2 jcos aj (18) the input data time step (i.e., 10 min, hourly (mid), hourly (int)
2 2
and daily data) while the rows present the output time horizon
and the anisotropic albedo formulation (r4) proposed by Arnfield (i.e., 10 min, hourly and daily). Such a configuration allows to
et al. [24]:
Table 10
Relative RMSE error index for 10 min, hourly and daily time step computed by the C
Table 8 model.
Values of anisotropic parameters in function of solar elevation angle used in Eq. (19)
taken from Arnfield (1975). RMSE [%] 10 min Hour Day
10 min 9.72 7.86 3.20
Parameters Morning Afternoon Hour mid e 8.84 13.01
a 0.244 0.212 Hour int e 9.54 7.16
b[
C1] 0.00526 0.00891 Day e e 18.75
720 C. Demain et al. / Renewable Energy 50 (2013) 710e721

Table 11 performed. Assuming the ground reflection process as being

Equations for parameters a, b, c, g, and h in Eq. (21). isotropic provides better results than considering non-isotropic
a ¼ sin dcos fsin bcos a  sin dsin fcos b approach. Daily and seasonally variation albedo formulations
b ¼ cos dcos fcos b þ cos dsin fsin bcos a minimize statistical error unlike a fixed albedo value.
c ¼ cos dsin asin b Finally, we have shown that the proposed model is able to
g ¼ sin dsin f
h ¼ cos dcos f
handle 10 min, hourly and daily input values. The best model
performance is obtained for hourly data. Nevertheless, our results
indicate that using a shorter input data time step than the output
Table 12 time horizon allows to reduce the RMSE.
Equations for parameters f1, f2, f3, f4 and f5 in Eq. (21).

f1 ¼ sin(uss)sin(usr) Acknowledgments
f2 ¼ cos(uss)cos(usr)
f3 ¼ ussusr This work was supported by the Belgian Science Policy Office
f4 ¼ sin(2uss)sin(2usr)
(BELSPO) through the ESA/PRODEX program PRODEX-9 contract No
f5 ¼ sin2(uss)sin2(usr)
4000102777 “Surface Solar Radiation”.

identify the best input data time step for a given output time Appendix
horizon. As an example, Table 10 indicates that using 10 min data
instead of daily data as input to generate daily integrated values To solve the daily integration of Eq. (3), the quantities cosqi and
reduce the RMSE from 18.75% to 3.20%. In general, whatever the cosqz need to be integrated from sunrise to sunset hour angle, i.e.,
output time horizon may be, the lower the input data time step is,
the better the result is. Using a shorter input data time step than the Zuss
output time horizon allows to smooth to some extend the model cos qi du ¼ sin dsin fcos bðuss  usr Þ
error (e.g. the RMSE decreases from 9.72 to 3.20% when integrating usr
the computed 10 min tilted global solar radiation over a day).
 sin dcos fsin bcos aðuss  usr Þ
Considering mid-hourly instantaneous values for the solar eleva-
tion angle (g) and the beam radiation incidence angle (qi) instead of þ cos dcos fcos bðsinðuss Þ  sinðusr ÞÞ
integrated hourly values do not significantly modify the coupled þ cos dsin fsin bcos aðsinðuss Þ  sinðusr ÞÞ
model performance on a hourly time horizon output (i.e. RMSE of
 cos dsin bsin aðcosðuss Þ  cosðusr ÞÞ (20)
8.84 and 9.54%, respectively). What is interesting to note here is
that for a similar input and output time horizon, the best model where usr and uss are the sunrise hour angle and sunset hour angle,
performances are obtained for hourly data. respectively.

6. Conclusion Zuss
ðbg  ahÞf1  cg f2 þ ð0:5bh  agÞf3 Þ
cos qz du ¼
bf1  cf2  af3
The amount of solar radiation available on a tilted surface is usr
a key factor in numerous solar energy applications, as thermal and 0:25bh f4 þ 0:5ch f5
photovoltaic energy systems or self-sustainable buildings. For þ (21)
bf1  cf2  af3
places where such solar data are not accessible, generating the
required data from conversion models would be an appropriate where parameters a, b, c, g and h are defined in Table 11 and f1, f2, f3,
alternative. A database consisting of solar horizontal global, normal f4 and f5 are defined in Table 12 (Allen et al. [27]).
incidence beam, horizontal diffuse and global solar radiation on In the case of a south-oriented panel with the slope equals to the
a south-oriented surface tilted at 50.79
recorded in Uccle (Brus- latitude: f ¼ b and a ¼ 0, Eqs. 20 and 21 can be rewritten as:
sels) during eight months (April to November 2011) have been used
Zuss
to determine the relative ability of 14 models to predict the global sinðuss Þ  sinðusr Þcos d
cos qi du ¼
solar radiation on a south-oriented tilted surface as a function of uss  usr
five different types of sky conditions. usr
Statistical evaluations revealed that models performance widely
Zuss
varies according to the sky conditions. Models discrepancies are bg f1 þ 0:5bh f3 þ 0:25bh f4
larger for intermediate sky conditions than for clear or overcast
cos qz du ¼ :
bf1
situation. It was observed that none of the considered transposition usr
model was able to score in each of the five sky categories. To
overcome such a limitation we have shown that a combination of References
Perez, Willmot and Bugler models lead to an improved estimation
of the global radiation on a south-oriented tilted surface. The Bugler [1] Liu B, Jordan R. Daily insolation on surfaces tilted towards the equator.
model [3] was found to perform best under all sky, clear and partly ASHRAE 1962;53:526e41.
[2] Willmot C. On the climatic optimization of the tilt and azimuth of flat-plate
clear conditions, whereas the Willmot model [2] was found to give solar collectors. Solar Energy 1982;28:205e16.
the best results under partly cloudy and cloudy conditions. Finally, [3] Bugler J. The determination of hourly insolation on an inclined plane using
Perez model [6] best fits measurements under overcast situation. a diffuse irradiance model based on hourly measured global horizontal
insolation. Solar Energy 1977;19:477e91.
The new coupled model provides relative RMSE in the range of
[4] Hay J. Study of shortwave radiation on non-horizontal surfaces. Downsview
7.22e21.62% (12e44.56 W m2) depending on the considered sky 1979;53. 79e12.
conditions. By comparison, the best individual model (Bu model) [5] Reindel D, Beckman W, Duffie J. Evaluation of hourly tilted surface radiation
generates an RMSE range of 5.65e22.71% (13.38e43.28 W m2). models. Renewable Energy 1990;45:9e17.
[6] Perez R, Seals R, Ineichen P, Stewart R, Menicucci D. A new simplified version
Sensitivity analysis of the new coupled model to the ground of the perez diffuse irradiance model for tilted surfaces. Solar Energy 1987;39:
reflected transposition factor as well as to the time integration was 221e31.
 C. Demain et al. / Renewable Energy 50 (2013) 710e721 721

[7] Gueymard C. An anisotropic solar irradiance model for tilted surfaces and its [25] Davies J, McKay D, Lunciani G, Abdel-Wahob M. Validation of models for
comparison with selected engineering algorithms. Solar Energy 1987;38: estimating solar radiation on horizontal surfaces. Atomspheric Environment
367e86. 1980:1.
[8] Noorian A, Moradi I, Kamali G. Evaluation of 12 models to estimate [26] Dickinson R. Land surface processes and climate surface albedo and energy
hourly diffuse irradiation on inclined surfaces. Renewable Energy 2008; balance. Advances in Geophysics 1983;25:305e53.
33:1406e12. [27] Allen R, Trezza R, Tasumi M. Analytical integrated functions for daily solar
[9] Gueymard C. Direct and indirect uncertainties in the prediction of tilted radiation on slopes. Energy and Environment 2010;1:547e54.
irradiance for solar engineering applications. Solar Energy 2009;83:432e44.
[10] Padovan A, Del Col D. Measurement and modeling of solar irradiance
components on horizontal and tilted planes. Solar Energy 2010;84:2068e84.
[11] Kambezidis H, Psiloglou B, Gueymard C. Measurements and models for total Nomenclature
solar irradiance on inclined surface in Athens, Greece. Solar Energy 1994;53:
177e85. So: Solar constant (1367 W m2)
[12] Perez R, Stewart R, Arbogast C, Seals R, Scott J. An anisotropic hourly diffuse Gext: Extraterrestrial solar irradiance (W m2)
radiation model for sloping surfaces: description, performance, validation, site BN: Direct beam solar irradiance (W m2)
dependency evaluation. Solar Energy 1986;36:481e97. G: Global horizontal solar irradiance (W m2)
[13] Hamilton H, Jackson A. A shield for obtaining diffuse sky radiation from Gcs: Global horizontal solar irradiance evaluated on clear sky condition (W m2)
portions of the sky. Solar Energy 1985;34:121e3. B: Direct horizontal solar irradiance (W m2)
[14] Korokanis P. On the choice of the angle of tilt for south facing solar collectors D: Diffuse horizontal solar irradiance (W m2)
in the Athens basin area. Solar Energy 1986;36:217e25. Gb: Global solar irradiance on inclined surface (W m2)
[15] Badescu V. 3d isotropic approximation for solar diffuse irradiance on tilted Bb: Direct solar irradiance on inclined surface (W m2)
surfaces. Renewable Energy 2002;26:221e3. Db: Diffuse solar irradiance on inclined surface (W m2)
[16] Ma C. Statistical comparison of models for estimating solar radiation on Db,iso: Isotropic diffuse component on inclined surface (W m2)
inclined surfaces. Solar Energy 1983;31:313e7. Db,cs: Circumsolar diffuse component on inclined surface (W m2)
[17] Perez R, Seals R, Ineichen P, Stewart R, Michalsky J. Modelling daylight Db,hb: Horizontal brightening component on inclined surface (W m2)
availability and irradiance components from direct and global irradiance. Rb: Reflected solar radiation on inclined surface (W m2)
Solar Energy 1990;44:271e89. Rd: Diffuse transposition factor
[18] Skartveit A, Olseth J. Modelling slope irradiance at high latitudes. Solar Energy Rr: Ground reflection transposition factor
1986;36:333e44. rb: Beam radiation conversion factor
[19] Muneer T. Solar radiation and daylight models for the energy efficient design kt: Clearness index
of buildings. Architectural press 1997; 26: 197e200. k0t : Modified clearness index
[20] Wlodarczyk D, Nowak H. Statistical analysis of solar radiation models onto M: Optical air mass
inclined planes for climatic conditions of lower Silesia in Poland. Archives of f: Latitude (rad)
Civil and Mechanical Engineering 2009;9:127e44. g: Solar altitude or elevation angle (rad)
[21] Evseev G, Kudsih A. The assessment of different models to predict the global b: Surface tilt angle (rad)
solar radiation on surface tilted to the south. Solar Energy 2009;83:377e88. d: Declination angle (rad)
[22] Temps R, Coulson K. Solar radiation incident upon slopes of different orien- u: Solar hour angle (rad)
tation. Solar Energy 1977;19:179e84. j: Solar azimuthal angle (rad)
[23] Klucher T. Evaluation of models to predict insolation on tilted surfaces. Solar qz: Solar zenital angle (rad)
Energy 1979;23:111e4. qi: Incidence angle on inclined surface (rad)
[24] Arnfield A. A note on the diurnal, latitudinal and seasonal variation of the a: Surface azimuthal angle (rad)
surface reflection coefficient. Journal of Applied Meteorology 1975;14:1603e8. r: Ground reflectance albedo