Académique Documents
Professionnel Documents
Culture Documents
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: An optimized articial neural network ensemble model is built to estimate daily global solar radiation
Received 21 May 2013 over large areas. The model uses clear-sky estimates and satellite images as input variables. Unlike most
Received in revised form studies using satellite imagery based on visible channels, our model also exploits all information within
2 September 2013
infrared channels of the Meteosat 9 satellite. A genetic algorithm is used to optimize selection of model
Accepted 5 September 2013
Available online 1 October 2013
inputs, for which twelve are selected e eleven 3-km Meteosat 9 channels and one clear-sky term. The
model is validated in Andalusia (Spain) from January 2008 through December 2008. Measured data from
83 stations across the region are used, 65 for training and 18 independent ones for testing the model. At
Keywords:
Global solar radiation
the latter stations, the ensemble model yields an overall root mean square error of 6.74% and correlation
Articial neural network coefcient of 99%; the generated estimates are relatively accurate and errors spatially uniform. The
Ensemble model model yields reliable results even on cloudy days, improving on current models based on satellite
Meteosat images imagery.
Satellite-derived irradiances 2013 Elsevier Ltd. All rights reserved.
Infrared channels
0360-5442/$ e see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.energy.2013.09.008
A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645 637
regions. In general, satellite count-to-irradiance models can be for photovoltaic systems. All these can be classied as empirical
grouped into physical and empirical models, but hybrid models are approaches, within the second type of scheme explained above.
the most extensively used. These models take advantage of both the In recent years, there have also been some works in global solar
former two types. That is, they have the robustness of a physical radiation estimation using ANNs and satellite data (e.g., for
approach (usually based on radiative transfer models that explicitly monthly [30], daily [22], and hourly [31] estimates). Usually, in
describe scattering and absorption processes operating in the these types of methods, satellite visible channels are used for
earth-atmosphere system), but simplify certain equations with model inputs. Only a few studies have used other satellite channels.
empirical regressions. There have been extensive reviews of these Perez et al. [14] suggested a method to improve the performance of
models [4e6]. Among the most used, we refer to Heliosat method satellite-to-irradiance models using satellite infrared sensors. Lu
[7e9] and Perez model [10e14] and their variations, some of which et al. [22] used instead all channels of a Multifunctional Transport
combine both model types. Although they retain some degree of Satellite of the Japan Meteorological Agency for daily global solar
tting to observations, their physical underpinnings guarantee estimation, in an effort to fully exploit information contained in
suitability for generalization. both visible and infrared channels. They achieved estimates with
It has been shown that these satellite-based approaches RMSE around 20%.
generate more accurate solar radiation estimates than classical In our study, Meteosat all-channel observations and a clear sky
interpolation methods, at locations outside the vicinity of a ground model were used as ANN inputs. The extra information from
station [10,15,16]. Examples of solar radiation maps built with infrared channels signicantly improves the derivation of solar
satellite data are found in Refs. [17,18]. radiation and yields more accurate estimates, outperforming other
In a previous work [19], we proposed a novel method within the physical or semi-empirical approaches.
third type of scheme. It estimates daily solar radiation at unob- The remainder of the paper is organized as follows. The Region
served locations, using reanalysis data from ECMWF (European and data section describes the study area and data, with special
Centre for Medium-Range Weather Forecasts) [20] as an alternative reference to Meteosat images. Detailed description of the method is
to satellite imagery. Despite the use of coarse spatial-resolution ERA provided in the following section, including neural network ar-
(ECMWF Reanalysis)-Interim data (0.7, about 70 km in the study chitecture and input selection. The Results and discussion section
region [20]), the method produced valid estimates of daily radia- deals with assessment of various ANN models. Finally, conclusions
tion, even better on a monthly basis. are given.
The aim of this work is to design a new method for spatial esti-
mation of daily solar irradiation via articial neural networks and 2. Region and data
satellite-derived irradiances, toward improving results in the current
literature on daily global solar radiation (e.g., RMSE (root mean square 2.1. Region of study and ground measurement dataset
error) between 9% and 16% on a daily basis, as cited in Refs. [8,21,22]).
Two main improvements are achieved herein: 1) An optimiza- The study region is Andalusia, which covers about 87,000 km2 in
tion procedure of an ANN (articial neural network) ensemble the southern Iberian Peninsula. The experimental dataset is the
model, and 2) use of additional information contained in Meteosat same used in a previous work [19]. Here, we consider data from the
channels other than the two visible ones, which are the most used year 2008. We used 83 solar radiation ground stations across the
in these methods. region. The stations are owned and maintained by the Andalusian
In recent years, ANN models have received increased scientic regional government, and are part of the Agricultural and Envi-
attention as a method for estimating solar radiation across large ronmental monitoring network. Two different pyranometers are
areas. These models for deriving solar radiation through ANNs use a used in this network: CMP3 (Kipp & Zonen), and SP1110 from
variety of meteorological variables as inputs. It has been shown Campbell Scientic (manufactured by Skye Instruments). Overall,
that ANN models outperform traditional regression approaches an instrumental error about 5% can be assumed [1,3].
[23e26], and hence have been used in numerous papers (see a brief Daily solar radiation values were carefully ltered to eliminate
review in Ref. [19]). We refer to Kalogirous reviews of ANN appli- spurious data, and inaccurate or inconsistent measurements [32].
cations for renewable energy systems [27,28], and Mellit et al. [29] Several quality control procedures were performed at this stage.
638 A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645
Particularly, a quality assessment based upon physical limits was effective radiances Cal Offset Cal Slope Pixel Count;
applied: an upper limit of 0.8 for the atmospheric clearness,
calculated as the daily measured solar radiation to the daily po-
(1)
tential extraterrestrial, and a lower limit of 0.01 MJ/m2 [33]. We only considered the rst eleven Meteosat-9 channels, with
As a rst step, 65 stations were selected for the training dataset, 3-km nadir spatial resolution (approximately 4.5 km at study area
and the remaining 18 were used as an independent testing dataset. latitudes). For these channels, a complete image, that is, the full
Fig. 1 shows the distribution of the training and validation stations disk of the earth, nominally consists of 3712 3712 pixels. The
within the study region. Next, 133 suspicious data were eliminated potential information contained in each channel may be summa-
from the training dataset. Ultimately, 23,592 daily global solar ra- rized as follows [38]:
diation values were used as the training dataset, and 6570 as the
testing dataset, corresponding to daily horizontal global solar - Channels 1e2 (VIS0.6 and VIS0.8): These are the visible chan-
irradiation observations from 2008/01/01 to 2008/12/31 at 65 nels, essential for cloud detection, cloud tracking, scene identi-
training stations (23,592 65 stations 365 days 133 suspicious cation, aerosol, and land surface and vegetation monitoring.
data) and 18 testing stations (6570 18 stations 365 days). - Channel 3 (NIR1.6): It can discriminate between snow and cloud,
ice and water clouds, and provides aerosol information.
2.2. Meteosat dataset - Channel 4 (IR3.9): Primarily for low cloud and fog detection.
Also supports measurement of land and sea surface temperature
2.2.1. MSG (Meteosat second generation) satellite series at night, and increases low-level wind coverage from cloud
The MSG satellite series, with its SEVIRI (Spinning Enhanced tracking. For MSG, the spectral band has been broadened to
Visible and Infrared Imager), scans the complete disk of the earth longer wavelengths for improving signal-to-noise ratio.
four times per hour and has 12 spectral channels. These are three - Channels 5e6 (WV6.2 and WV7.3): Channel for observing water
solar channels (0.6, 0.8 and 1.6 mm), eight infrared channels (3.9, 6.2, vapour and winds. Enhanced to two channels peaking at
7.3, 8.7, 9.7, 10.8, 12.0 and 13.4 mm), and one high-resolution broad- different levels in the troposphere. Also supports height allo-
band visible channel (0.3e0.7 mm). The nadir spatial resolution of cation of semitransparent clouds.
SEVIRI is 1 1 km2 for the high-resolution channel, and 3 3 km2 - Channel 7 (IR8.7): Provides quantitative information on thin
for all other channels [34,35]. Raw data are internally managed via a cirrus clouds and supports discrimination between ice and
radiometric process, in four main steps [34]: a) linearization, b) water clouds.
conversion into radiances, c) calibration, and d) linear scaling. The - Channel 8 (IR9.7): Ozone radiances may be used as an input to
last step ensures that the necessary dynamic range falls into the numerical weather prediction. Temporal evolution of the total
available interval [0, 1023], through two linear coefcients: Cal_- ozone eld can also be monitored.
Slope and Cal_Offset. Both are constants provided for each image - Channels 9e10 (IR10.8 and IR12.0): Well-known, split-window
by the satellite sensors. Finally, effective radiances for each channel channels. Essential for measuring sea and land surface and
are fully dened by the relationship [35e37]: cloud-top temperatures.
Fig. 1. Study region. Highlighted region shows DEM (digital model of terrain) used.
A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645 639
predetermined number of epochs, set to 500. In this case, minimum and clear-sky term, and the second with all possible inputs except
error was achieved before reaching the maximum epoch. channel 7, which is the term with lowest relative importance.
Table 1 shows all relative importances of terms, computed as the
3.2. Selection of inputs sum of relative evidence weights of all possible combinations in
which the term appears. Despite similar AIC scores of the two best
One of the more critical steps in ANN modelling is selection of an input combinations, the glmulti algorithm gives a very different
appropriate set of input variables [47e49]. If relevant inputs are relative weight to each candidate. Ranked relative evidence
omitted, the model is unable to capture the desired inputeoutput weights of the models are computed as exp(DAIC/2), where DAIC is
relationships. On the other hand, too many inputs can produce the difference of AIC between a model and the best model. These
over-parameterization problems, as cited above. Usually, a set of weights are normalized so that their sum is one; they can be
candidate input variables is found via a priori knowledge of the interpreted as probabilities of each model being the best in the set
system being modelled. [55]. The relative weight computed for the rst combination (all
In this work, the candidate inputs were a clear-sky solar radia- possible inputs) is 90.3%, against 9.70% for the other combination
tion estimate and the eleven 3-km Meteosat channels. These sat- (all inputs except channel 7). Hence, the best combination was one
ellite observations provide useful information of scattering, built with all possible candidate inputs, i.e., with no input
absorption and reection processes affecting solar radiation pass- neglected.
ing through the atmosphere.
n
1 X b
MBE Gd G d (4)
n
d1
v
u n 2
uX
RMSE t Gd G b =n (5)
d1
r
X 2 X 2
R bs G =
G Gs G ; (7)
Fig. 4. Comparison between measured and predicted radiation values, grouped by training stations.
other interpolation techniques such as kriging, which always re- RoeSevilla) had slightly worse error estimates than other stations.
quires real values. This fact demonstrates the generalization capability of the model,
To certify the spatial interpolation capability of the ANN-E and allows use of the ANN-E model to generate reliable estimates at
model, we checked the accuracy of estimates at each station. any unobserved point within the study region, thereby generating
Figs. 4 and 5 show statistical scores of radiation estimates at each of accurate daily solar radiation maps.
the ground training and independent stations, respectively. Nearly
all the stations have relative RMSE values between 4% and 9%. A 4.2. Dependence on sky conditions
boxplot (Fig. 6) shows overall scores. Within each box, the central
mark is the median of all the error values, and box edges denote the Fig. 7 shows error values of estimates grouped in quarters of a
25th and 75th percentiles. Whiskers extend to the most extreme year. In Andalusia, summer is usually sunny and dry, and winter is
data points not considered outliers, and outliers are plotted cloudy and rainy. Hence, from April through September there are
individually. frequent clear-sky days, and the ANN-E model yields more accurate
These results conrm that the ANN-E model gave reliable results estimates than those from rainy months.
over the entire study region, with accurate estimates and spatially To better evaluate the model for cloudy days, we calculated a
uniform errors. Only three training stations of the 65 (6th station: clear-sky index KC. It is dened as the ratio of measured global
La MojoneraeAlmera; 41st: HuesaeJan, and 51st: Esteponae irradiance GD to that received under clear-sky conditions, GCS,
Mlaga) and one independent station of the 18 (17th: Puebla del which is estimated with the ESRA clear-sky solar radiation model
Fig. 5. Comparison between measured and predicted radiation values, grouped by testing stations.
A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645 643
Fig. 7. Error values of estimates grouped in quarters of a year, for a) training and b) testing stations.
644 A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645
Table 3
Error values of ANN-E model, grouped by clear-sky index values.
KC < 0.5 0.5 KC 0.8 KC > 0.8 KC < 0.5 0.5 KC 0.8 KC > 0.8
to obtain more accurate solar radiation estimates every 15 min, by [9] Zarzalejo LF, Polo J, Martn L, Ramrez L, Espinar B. A new statistical approach
for deriving global solar radiation from satellite images. Sol Energy
using all information contained in Meteosat channels.
2009;83(4):480e4.
[10] Perez R, Seals R, Zelenka A. Comparing satellite remote sensing and ground
network measurements for the production of site/time specic irradiance
Acknowledgements data. Sol Energy 1997;60(2):89e96.
[11] Ineichen P, Perez R. Derivation of cloud index from geostationary satellites
and application to the production of solar irradiance and daylight illuminance
This work was supported by the Spanish Ministry of Science and
data. Theor Appl Climatol 1999;64(1e2):119e30.
Technology (Project CGL2011-30377-C02-01). The data were kindly [12] Perez R, Ineichen P, Moore K, Kmiecik M, Chain C, George R, et al. A new
provided by the regional ofce of Agriculture and Fishing of operational model for satellite-derived irradiances: description and valida-
Andalusia. tion. Sol Energy 2002;73(5):307e17.
[13] Cebecauer T, Suri M, Perez R. High performance MSG satellite model for
operational solar energy applications. In: Proc ASES annual conference,
Phoenix, AZ 2010.
References [14] Perez R, Kivalov S, Zelenka A, Schlemmer J, Hemker Jr K. Improving the per-
formance of satellite-to-irradiance models using the satellites infrared sen-
[1] Alsamamra H, Ruiz-Arias JA, Pozo-Vzquez D, Tovar-Pescador J. A comparative sors. In: Proc ASES annual conference, Phoenix, Arizona 2010.
study of ordinary and residual kriging techniques for mapping global solar [15] Zelenka A, Czeplak G, dAgostino V, Josefson W, Maxwell E, Perez R. Tech-
radiation over southern Spain. Agric For Meteorol 2009;149(8):1343e57. niques for supplementing solar radiation network data. Technical report.
[2] Ruiz-Arias JA, Tovar-Pescador J, Pozo-Vazquez D, Alsamamra H. A comparative Krahbuhlstrasse, 58, CH-8044 Zurich, Switzerland: International Energy
analysis of DEM-based models to estimate the solar radiation in mountainous Agency, # IEA-SHCP-9D-1, Swiss Meteorological Institute; 1992.
terrain. Int J Geogr Inf Sci 2009;23(8):1049e76. [16] Zelenka A, Perez R, Seals R, Renn D. Effective accuracy of satellite-derived
[3] Ruiz-Arias JA, Pozo-Vzquez D, Santos-Alamillos FJ, Lara-Fanego V, Tovar- hourly irradiances. Theor Appl Climatol 1999;62(3):199e207.
Pescador J. A topographic geostatistical approach for mapping monthly mean [17] Janjai S, Laksanaboonsong J, Nunez M, Thongsathitya A. Development of a
values of daily global solar radiation: a case study in southern Spain. Agric For method for generating operational solar radiation maps from satellite data for
Meteorol 2011;151(12):1812e22. a tropical environment. Sol Energy 2005;78(6):739e51.
[4] Noia M, Ratto CF, Festa R. Solar irradiance estimation from geostationary [18] Janjai S, Pankaew P, Laksanaboonsong J, Kitichantaropas P. Estimation of solar
satellite data: I. Statistical models. Sol Energy 1993;51(6):449e56. radiation over Cambodia from long-term satellite data. Renew Energy
[5] Noia M, Ratto CF, Festa R. Solar irradiance estimation from geostationary 2011;36(4):1214e20.
satellite data: II. Physical models. Sol Energy 1993;51(6):457e65. [19] Linares-Rodrguez A, Ruiz-Arias JA, Pozo-Vzquez D, Tovar-Pescador J. Gen-
[6] Pinker RT, Frouin R, Li Z. A review of satellite methods to derive surface eration of synthetic daily global solar radiation data based on ERA-Interim
shortwave irradiance. Remote Sens Environ 1995;51(1):108e24. reanalysis and articial neural networks. Energy 2011;36(8):5356e65.
[7] Cano D, Monget JM, Aubuisson M, Guillard H, Regas N, Wald L. A method for [20] Simmons A, Uppala S, Dee D, Kobayashi S. ERA-Interim: new ECMWF rean-
the determination of global solar radiation from meteorological satellite data. alysis products from 1989 onwards. ECMWF Newsletter No 11; 2007.
Sol Energy 1986;37:31e9. [21] Benghanem M, Mellit A, Alamri SN. ANN-based modelling and estimation of
[8] Rigollier C. The method Heliosat-2 for deriving shortwave solar radiation from daily global solar radiation data: a case study. Energy Convers Manag
satellite images. Sol Energy 2004;77(2):159e69. 2009;50(7):1644e55.
A. Linares-Rodriguez et al. / Energy 61 (2013) 636e645 645
[22] Lu N, Qin J, Yang K, Sun J. A simple and efcient algorithm to estimate daily global [46] Dorvlo ASS, Jervase JA, Al-Lawati A. Solar radiation estimation using articial
solar radiation from geostationary satellite data. Energy 2011;36(5):3179e88. neural networks. Appl Energy 2002;71:307e19.
[23] Reddy K. Solar resource estimation using articial neural networks and [47] Lopez G, Batlles FJ, Tovar-Pescador J. Selection of input parameters to model
comparison with other correlation models. Energy Convers Manag direct solar irradiance by using articial neural networks. Energy 2005;30(9):
2003;44(15):2519e30. 1675e84.
[24] Tymvios F, Jacovides C, Michaelides S, Scouteli C. Comparative study of Ang- [48] Behrang MA, Assareh E, Ghanbarzadeh A, Noghrehabadi AR. The potential of
strms and articial neural networks methodologies in estimating global different articial neural network (ANN) techniques in daily global solar ra-
solar radiation. Sol Energy 2005;78(6):752e62. diation modeling based on meteorological data. Sol Energy 2010;84(8):1468e
[25] Elminir H, Azzam Y, Younes F. Prediction of hourly and daily diffuse fraction 80.
using neural network, as compared to linear regression models. Energy [49] Koca A, Oztop HF, Varol Y, Koca GO. Estimation of solar radiation using arti-
2007;32(8):1513e23. cial neural networks with different input parameters for Mediterranean
[26] Jiang Y. Computation of monthly mean daily global solar radiation in China region of Anatolia in Turkey. Expert Syst Appl 2011;38(7):8756e62.
using articial neural networks and comparison with other empirical models. [50] Rigollier C, Bauer O, Wald L. On the clear sky model of the ESRA European
Energy 2009;34(9):1276e83. solar radiation atlas with respect to the Heliosat method. Sol Energy
[27] Kalogirou SA. Applications of articial neural networks for energy systems. 2000;68(1):33e48.
Appl Energy 2000;67:17e35. [51] Batlles F, Bosch J, Tovar-Pescador J, Martinez Durban M, Ortega R, Miralles I.
[28] Kalogirou SA. Articial neural networks in renewable energy systems appli- Determination of atmospheric parameters to estimate global radiation in
cations: a review. Renew Sustain Energy Rev 2001;5:373e401. areas of complex topography: generation of global irradiation map. Energy
[29] Mellit A, Kalogirou S. Articial intelligence techniques for photovoltaic ap- Convers Manag 2008;49(2):336e45.
plications: a review. Prog Energy Combust Sci 2008;34(5):574e632. [52] Ineichen P. Comparison of eight clear sky broadband models against 16 in-
[30] Senkal O. Modeling of solar radiation using remote sensing and articial dependent data banks. Sol Energy 2006;80(4):468e78.
neural network in Turkey. Energy 2010;35(12):4795e801. [53] Bosch J, Lopez G, Batlles F. Daily solar irradiation estimation over a moun-
[31] Zarzalejo L. Articial intelligence techniques applied to hourly global irradiance tainous area using articial neural networks. Renew Energy 2008;33(7):
estimation from satellite-derived cloud index. Energy 2005;30(9):1685e97. 1622e8.
[32] Polo J, Zarzalejo L, Ramirez L, Espinar B. Iterative ltering of ground data for [54] Yacef R, Benghanem M, Mellit A. Prediction of daily global solar irradiation
qualifying statistical models for solar irradiance estimation from satellite data. data using Bayesian neural network: a comparative study. Renew Energy
Sol Energy 2006;80(3):240e7. 2012;48(0):146e54.
[33] Iqbal M. An introduction to solar radiation. Toronto; New York: Academic [55] Calcagno V, de Mazancourt C. glmulti: an R package for easy automated model
Press; 1983. selection with (generalized) linear models. J Stat Softw 2010;34(12).
[34] MSG level 1.5 image data format description. EUM/MSG/ICD/105; 2005 (3). [56] Akaike H. A new look at the statistical model identication. Autom Control
[35] Effective radiance and brightness temperature relation for Meteosat-8 and 9. IEEE Trans 1974;19(6):716e23.
EUM/OPS-MSG/TEN/08/0024http://www.eumetsat.int; 2008. [57] Goldberg DE. Genetic algorithms in search, optimization and machine
[36] A planned change to the MSG level 1.5 image product radiance denition, learning. Addison-Wesley Longman Publishing Co., Inc.; 1989.
v1A. EUM/OPS-MSG/TEN/06/0519; 2007. [58] Trevino V, Falciani F. GALGO: an R package for multivariate variable selection
[37] Radiometric calibration of MSG SEVIRI level 1.5 image data in equivalent using genetic algorithms. Bioinformatics 2006;22(9):1154e6.
spectral blackbody radiance. EUM/OPS-MSG/TEN/03/0064; 2007. [59] Sharkey AJC. Combining articial neural nets: ensemble and modular multi-
[38] Schmetz J, Pili P, Tjemkes S, Just D, Kerkmann J, Rota S, et al. An introduction to net systems. New York: Springer-Verlag; 1999.
Meteosat second generation (MSG). Bull Am Meteorol Soc 2002;83(7):977e92. [60] Estvez J, Gaviln P, Girldez JV. Guidelines on validation procedures for
[39] https://eoportal.eumetsat.int. [accessed 12.07.13]. meteorological data from automatic weather stations. J Hydrol 2011;402(1e
[40] http://code.google.com/p/nctoolbox/. [accessed 12.07.13]. 2):144e54.
[41] Bishop CM. Neural networks for pattern recognition. Oxford University Press, [61] Mefti A, Adane A, Bouroubi MY. Satellite approach based on cloud cover
Inc.; 1995. classication: estimation of hourly global solar radiation from Meteosat im-
[42] The MathWorks, Inc., 3 Apple Hill Drive, Natick, MA 01760-02098, USA. ages. Energy Convers Manag 2008;49(4):652e9.
[43] Hagan MT, Demuth HB, Beale MH. Neural network design. Boston, MA: PWS [62] Roerink GJ, Bojanowski JS, de Wit AJW, Eerens H, Supit I, Leo O, et al. Evalu-
Publishing; 1996. ation of MSG-derived global radiation estimates for application in a regional
[44] MacKay DJC. Bayesian interpolation. Neural Comput 1992;4(3):415e47. crop model. Agric For Meteorol 2012;160:36e47.
[45] Foresee FD, Hagan MT. Gauss-Newton approximation to Bayesian regulari- [63] Bosch JL, Batlles FJ, Zarzalejo LF, Lpez G. Solar resources estimation
zation. In: Proceedings of the 1997 international joint conference on neural combining digital terrain models and satellite images techniques. Renew
networks 1997. p. 1930e5. Energy 2010;35(12):2853e61.