You are on page 1of 7

Global solar radiation estimation using sunshine

duration in Spain
J. Almorox
*
, C. Hontoria
Departamento de Edafologa, Escuela T ecnica Superior de Ingenieros Agr onomos, Universidad Polit ecnica de Madrid,
Avd. Complutense s/n, 28040 Madrid, Spain
Received 17 April 2003; accepted 19 August 2003
Abstract
Several equations were employed to estimate global solar radiation from sunshine hours for 16 mete-
orological stations in Spain, using only the relative duration of sunshine. These equations included the
original Angstr omPrescott linear regression and modied functions (quadratic, third degree, logarithmic
and exponential functions). Estimated values were compared with measured values in terms of the coe-
cient of determination, standard error of the estimate and mean absolute error. All the models tted the
data adequately and can be used to estimate global solar radiation from sunshine hours. This study nds
that the third degree models performed better than the other models, but the linear model is preferred due
to its greater simplicity and wider application. It is also found that seasonal partitioning does not signi-
cantly improve the estimation of global radiation.
2003 Elsevier Ltd. All rights reserved.
Keywords: Angstr om equation; Global solar radiation estimation; Sunshine duration; Correlation models
1. Introduction
Knowledge of local solar radiation is essential for many applications, including architectural
design, solar energy systems and particularly for design methods, crop growth models and
evapotranspiration estimates in the design of irrigation systems. In spite of the importance of
solar radiation measurements, this information is not readily available due to the cost and
maintenance and calibration requirements of the measuring equipment. The limited coverage of
radiation values dictates the need to develop models to estimate solar radiation based on other,
Energy Conversion and Management 45 (2004) 15291535
www.elsevier.com/locate/enconman
*
Corresponding author. Tel.: +34-91-3365-689; fax: +34-91-3365-680.
E-mail address: jaal@eda.etsia.upm.es (J. Almorox).
0196-8904/$ - see front matter 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enconman.2003.08.022
more readily available, data [1]. Several empirical models have been used to calculate solar ra-
diation, using variables such as sunshine hours [2], air temperature [3], precipitation [4], relative
humidity [5] and cloudiness [6]. The most commonly used parameter for estimating global solar
radiation is sunshine duration. Sunshine duration can be easily and reliably measured and data
are widely available. Most of the models for estimating solar radiation that appear in the liter-
ature use the sunshine ratio [1]. The most widely used method is that of Angstr om [2], who
proposed a linear relationship between the ratio of average daily global radiation to the corre-
sponding value on a completely clear day and the ratio of average daily sunshine duration to the
maximum possible sunshine duration. The problem of determining clear sky global irradiance was
bypassed by Prescott, who suggested using extraterrestrial radiation intensity values instead.
Prescott [7] put the equation in a more convenient form by replacing the average global radiation
on a clear day with the extraterrestrial radiation.
The objective of this study was to validate several expression models for the prediction of
monthly average daily global radiation on a horizontal surface from sunshine hours and to select
the most adequate model.
2. Materials and methods
The Angstr omPrescott formula is
H=H
0
a bn=N 1
where H and H
0
are, respectively, the monthly mean daily global radiation and the daily extra-
terrestrial radiation on a horizontal surface; n and N are, respectively, the monthly average daily
sunshine duration and the monthly average maximum possible daily sunshine duration; and a and
b are empirically determined regression constants.
The sky is overcast when the ratio n=N is zero, so the regression coecient a is a measure of
the overall atmospheric transmission for total cloud conditions, while the term b is the rate of
increase of H=H
0
with n=N. The sum a b denotes the overall atmospheric transmission under
clear sky conditions. The mean transmission has a typical factor value of around 0.75 [8]. These
constants can assume a wide range of values, depending on the location considered, and can be
inferred from correlations established at neighboring locations. Thus, for example, when using the
FAO 56 Penman estimation method [9], Angstr om values calibration is recommended. Variations
in the a and b values are explained as a consequence of local and seasonal changes in the type and
thickness of cloud cover, the eects of snow covered surfaces, the concentrations of pollutants and
latitude [8,10].
Several types of regression models have been proposed in the literature for estimating global
solar radiation from extraterrestrial irradiance and measured and theoretical daily sunshine
duration (Table 1).
The daily extraterrestrial radiation on a horizontal surface at the middle of the month can be
calculated as a function of the solar constant (G
sc
), the eccentricity correction factor of the Earths
orbit (E
0
), the latitude of the site (U), the solar declination (d) and the mean sunrise hour angle
(w
s
) in MJ m
2
day
1
using the following equation:
1530 J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535
H
0
1=p G
sc
E
0
cos U cos d sinw
s
p=180 sin U sind w
s
2
The solar constant is the amount of energy received at the top of the Earths atmosphere,
measured at an average distance between the Earth and Sun on a surface oriented perpendicular
to the Sun. The generally accepted solar constant of 1367 Wm
2
(an equivalent daily value of
118.108 MJ m
2
day
1
to reconcile units) is a satellite measured yearly average.
The eccentricity correction factor can be calculated by the expression [14]:
E
0
1:00011 0:034221 cos C 0:00128 sin C 0:000719 cos 2C 0:000077 sin2C
3
The solar declination d can be computed in degrees from the equation [14]:
d 180=p 0:006918 0:399912 cos C 0:070257 sinC 0:006758 cos 2C
0:000907 sin 2C 0:002697 cos 3C 0:00148 sin 3C 4
where C 2p n 1=365 (radians) and n is the number of the day of the year, starting from the
rst of January.
The geometric mean sunrise hour angle for the month on a horizontal surface can be calculated
(except in the absence of sunrise and sunset, during the polar day and night) in degrees from:
w
s
cos
1
tan U tand 5
The maximum possible number of daylight hours is given by:
N 2=15 w
s
6
The main purpose of this work was to obtain sets of regression constants for ve dierent re-
gression models for as many weather stations as possible with data available in Spain. To this end,
we employed the measured data of monthly average global solar radiation on horizontal surfaces
and sunshine hours from 16 meteorological stations across Spain, given in Table 2, to generate the
equations. The data were obtained from the National Meteorological Institute (Ministerio de
Medio Ambiente, Spain).
Five dierent regression models were applied, four proposed in the literature (linear, quadratic,
polynomial third degree and logarithmic) and an exponential model used in this work. Coecient
values were calculated from regression analysis between H=H
0
and n=N for a long period and each
month. This method is considered by Tadros [15] as the best for predicting global solar radiation
over eight meteorological stations in Egypt. To determine the annual regressions, we used data
from all months of the year. Since coecients vary from season to season [16,17], the data was
Table 1
Regression models proposed in the literature
Models Regression equations Source
Linear H=H
0
a b n=N Angstr omPrescott [2,7]
Quadratic H=H
0
a b n=N c n=N
2
Akinoglu and Ecevit [11]
Third degree H=H
0
a b n=N c n=N
2
d n=N
3
Ertekin and Yaldiz [12]
Logarithmic H=H
0
a b logn=N Ampratwum and Dorvlo [13]
J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535 1531
separated into the summer (June, July, August), autumn (September, October, November), spring
(March, April, May) and winter (December, January, February) seasons.
The regression models were developed to nd the best predictive equation using Statgraphics
Plus (version 5) software. The accuracy of the estimated values was tested by calculating the R-
squared statistic, the standard error of estimate and the mean absolute error. The R-squared
statistic indicates the percentage of variability of the dependent variable as it was tted in the full
model. The standard error of the estimate shows the value for the standard deviation of the re-
siduals, estimated by the square root of the mean squared error. The mean absolute error is a
measure of the forecast accuracy, calculated by summing the absolute errors and dividing by the
number of observations. The square root of the mean squared error and the mean absolute error
are the fundamental measures of accuracy [18].
3. Results and discussion
The equations developed and the R
2
, SEE and MAE values of the equations are given in Tables
3 and 4. The model results for annual data are listed in Table 3, and the results of the regression
analyses for winter, spring, summer and autumn can be seen in Table 4.
Comparing the results, we can see that all the regression equations gave very good results. The
quadratic model and the third degree equation gave the best estimate and have the smallest errors
for the annual and seasonal values. The logarithmic and exponential models performed worse
than the other models, giving the largest standard error of the estimate and mean absolute error.
The linear, second and third degree regression equations gave good, and very similar, accuracy.
Given the small dierences between the variance explained by linear regression and the most
Table 2
Geographical location and period of data of the meteorological stations used
Station Latitude (N) Longitude Elevation (m) Period of observation
Badajoz 38 53
0
00
00
6 49
0
45
00
W 185 19761984
Bilbao 43 17
0
53
00
2 54
0
21
00
W 39 19852000
Caceres 39 28
0
20
00
6 20
0
22
00
W 405 19832000
Cadiz 36 29
0
55
00
6 15
0
37
00
W 8 19912000
Castell on 39 57
0
00
00
0 01
0
00
00
W 35 19841991
La Coru~ na 43 22
0
02
00
8 25
0
10
00
W 58 19852000
Logro~ no 42 27
0
50
00
2 22
0
47
00
W 364 19831998
Madrid 40 27
0
10
00
3 43
0
27
00
W 664 19761999
Murcia 38 00
0
10
00
1 10
0
10
00
W 62 19852000
Oviedo 43 21
0
13
00
5 52
0
24
00
W 336 19811999
San Sebastian 43 18
0
24
00
2 02
0
22
00
W 259 19832000
Santander 43 27
0
53
00
3 49
0
08
00
W 64 19831998
Sevilla 37 25
0
15
00
5 53
0
47
00
W 34 19811988
Toledo 39 53
0
05
00
4 02
0
58
00
W 516 19831998
Tortosa 40 49
0
14
00
0 29
0
29
00
E 48 19801998
Valladolid 41 42
0
00
00
4 51
0
00
00
W 846 19821998
Source: Ministerio de Medio Ambiente, Spain.
1532 J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535
accurate equation, the third degree one (86.4% and 86.5%, for the annual model), and the sim-
plicity and wider application of the linear equation, the simple linear regression, is in practice,
sucient, and the use of the quadratic and third degree equations is not suciently justied.
Table 3
Annual regression results, and R
2
, SEE, and MAE values of models
Equations R
2
SEE MAE
Annual
Linear H=H
0
0:2170 0:5453 n=N 0.864 0.0361 0.0271
Quadratic H=H
0
0:1840 0:6792 n=N 0:1228 n=N
2
0.865 0.0359 0.0270
Third degree H=H
0
0:230 0:3809 n=N 0:4694 n=N
2
0:3657 n=N
3
0.865 0.0359 0.0269
Logarithmic H=H
0
0:6902 0:6142 logn=N 0.842 0.0388 0.0298
Exponential H=H
0
0:0271 0:3096 e
n=N
0.851 0.0378 0.0286
R
2
(R-squared): coecient of determination.
SEE: standard error of the estimate.
MAE: mean absolute error.
Table 4
Regression results for winter, spring, summer and autumn, and R
2
(R-squared), SEE (standard error of the estimate),
and MAE (mean absolute error) values
Equations R
2
SEE MAE
Winter
Linear H=H
0
0:2114 0:5291 n=N 0.865 0.0314 0.0235
Quadratic H=H
0
0:1875 0:6372 n=N 0:11054 n=N
2
0.866 0.0313 0.0235
Third degree H=H
0
0:241 0:2496 n=N 0:7355 n=N
2
0:5709 n=N
3
0.867 0.0235 0.0235
Logarithmic H=H
0
0:6458 0:5280 logn=N 0.839 0.0344 0.0262
Exponential H=H
0
0:0549 0:3181 e
n=N
0.853 0.0328 0.0245
Spring
Linear H=H
0
0:2414 0:5375 n=N 0.825 0.0351 0.0250
Quadratic H=H
0
0:2216 0:6188 n=N 0:0773 n=N
2
0.826 0.0351 0.0249
Third degree H=H
0
0:2306 0:5590 n=N 0:0455 n=N
2
0:0792 n=N
3
0.826 0.0351 0.0249
Logarithmic H=H
0
0:6993 0:5889 logn=N 0.803 0.0372 0.0265
Exponential H=H
0
0:0105 0:3128 e
n=N
0.817 0.0359 0.0260
Summer
Linear H=H
0
0:2518 0:5004 n=N 0.876 0.0334 0.0246
Quadratic H=H
0
0:1967 0:7040 n=N 0:1705 n=N
2
0.878 0.0331 0.0243
Third degree H=H
0
0:2868 0:1725 n=N 0:8010 n=N
2
0:5579 n=N
3
0.879 0.0330 0.0241
Logarithmic H=H
0
0:7074 0:6327 logn=N 0.865 0.0348 0.0259
Exponential H=H
0
0:0517 0:2702 e
n=N
0.863 0.0350 0.0263
Autumn
Linear H=H
0
0:1966 0:5580 n=N 0.870 0.0318 0.0235
Quadratic H=H
0
0:1755 0:6465 n=N 0:0853 n=N
2
0.870 0.0318 0.0235
Third degree H=H
0
0:2733 0:0063 n=N 1:2665 n=N
2
0:8793 n=N
3
0.872 0.0316 0.0234
Logarithmic H=H
0
0:6717 0:6081 logn=N 0.849 0.0342 0.0263
Exponential H=H
0
0:0694 0:3271 e
n=N
0.862 0.0328 0.0247
J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535 1533
From the results, it is found that the correlations in summer and autumn have better perfor-
mance than in spring, and the annual correlation has a similar performance to the winter cor-
relation but better than that of the spring. The results also show that, in practice, the annual linear
model is sucient and that the seasonal partitioning of data is not justied.
According to our results, the linear annual equation proposed for the estimation of global solar
radiation for Spain is
H=H
0
0:2170 0:5453 n=N 7
The values for the coecients of the linear equation are quite consistent with the values reported
in the literature. For example, Louche et al. [19] reported the values of 0.206 and 0.546 for a
French Mediterranean site; FAO 56 [9] recommended, where no actual solar radiation data are
available, the values of 0.25 and 0.50. Similar results have also been reported in the CGMS model
[20] for Spain, 0.253 and 0.502.
4. Conclusion
Several sunshine based models have been employed for estimating global solar radiation for
Spain. The dierences between the results of the dierent models are negligible. The third degree
equation is the best overall, according to R-squared, SEE and MAE, and has the best performance
based on the measured data at 16 stations in Spain. We recommend the use of the linear equation,
however, due to the small dierences between the linear regression and the third degree one and
the greater simplicity of the linear equation. It is found that the seasonal partitioning of data is
not justied. The linear annual equation can be used to estimate the global solar radiation in
Spain.
References
[1] Al-Lawati A, Dorvlo ASS, Jervase JA. Monthly average daily solar radiation and clearness index contour maps
over Oman. Energy Conversion and Management 2003;44:691705.
[2] Angstr om A. Solar and terrestrial radiation. Quart J Roy Met Soc 1924;50:1215.
[3] Hargreaves GH, Samani ZA. Estimating potential evapotranspiration. J Irrigat Drainage Div, ASCE 1982;
108(1 IR3):22330.
[4] De Jong R, Stewart DW. Estimating global radiation from common meteorological variables in western Canada.
Can J Plant Sci 1993;73:50918.
[5] Elagib NA, Babiker SF, Alvi SH. New empirical models for global solar radiation over Bahrain. Energy Conver
Mgmt 1998;39(8):82735.
[6] Black JN. The distribution of solar radiation over the earths surface. Arch Met Geoph Biokl 1956;7:16589.
[7] Prescott JA. Evaporation from a water surface in relation to solar radiation. Trans R Soc Sci Austr 1940;64:114
25.
[8] Linacre E. Climate data and resources. A reference and guide. London: Routledge; 1992. p. 366.
[9] Allen RG, Pereira LS, Raes D, Smith M. Crop evapotranspiration. Guidelines for computing crop water
requirements. FAO Irrigation and Drainage paper 56. Rome, 1998.
[10] Boisvert JB, Hayhoe HN, Dube PA. Improving the estimation of global solar radiation across Canada. Agric For
Meteorol 1990;52:27586.
1534 J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535
[11] Akinoglu BG, Ecevit A. Construction of a quadratic model using modied Angstr om coecients to estimate global
solar radiation. Solar Energy 1990;45:8592.
[12] Ertekin C, Yaldiz O. Comparison of some existing models for estimating global solar radiation for Antalya
(Turkey). Energy Convers Mgmt 2000;41:31130.
[13] Ampratwum DB, Dorvlo ASS. Estimation of solar radiation from the number of sunshine hours. Appl Energy
1990;62:1617.
[14] Spencer JW. Fourier series representation of the position of the Sun. Search 1971;2(5):172.
[15] Tadros MTY. Uses of sunshine duration to estimate the global solar radiation over eight meteorological stations in
Egypt. Renew Energy 2000;21:23146.
[16] Hussain M, Rahman L, Rahman MM. Techniques to obtain improved predictions of global radiation from
sunshine duration. Renew Energy 1999;18:26375.
[17] Togrul IT, Togrul H, Evin D. Estimation of monthly global solar radiation from sunshine duration measurement
in Elazig. Renew Energy 2000;19:58795.
[18] Chegaar M, Chibani A. Global solar radiation estimation in Algeria. Energy Conver Mgmt 2001;42:96773.
[19] Louche A, Notton G, Poggi P, Simonnot G. Correlations for direct normal and global horizontal irradiations on a
French Mediterranean site. Solar Energy 1991;46(4):2616.
[20] Supit I, Van der Goot E, editors. Updated system description of the WOFOST crop growth simulation model as
implemented in the crop growth monitoring system applied by the European Commission. Internet On-line Book.
Trebook 7. Heelsum, The Netherlands: Treemail Publishers; 2003.
J. Almorox, C. Hontoria / Energy Conversion and Management 45 (2004) 15291535 1535

You might also like