Sie sind auf Seite 1von 20

(This is a sample cover image for this issue. The actual cover is not yet available at this time.

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elseviers archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Author's personal copy

Renewable Energy 55 (2013) 85e103

Contents lists available at SciVerse ScienceDirect

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

Accuracy analysis for fty-four clear-sky solar radiation models using routine hourly global irradiance measurements in Romania
Viorel Badescu a, b, *, Christian A. Gueymard c, Sorin Cheval d, Cristian Oprea e, Madalina Baciu e, Alexandru Dumitrescu e, f, Flavius Iacobescu a, Ioan Milos e, Costel Rada e
a

Candida Oancea Institute, Polytechnic University of Bucharest, Spl. Independentei 313, Bucharest 060042, Romania Romanian Academy, Calea Victoriei 125, Bucharest, Romania Solar Consulting Services, P.O. Box 392, Colebrook, NH 03576, USA d National Research and Development Institute for Environmental Protection, Splaiul Independentei nr. 294, Sect. 6, 060031 Bucuresti, Romania e National Meteorological Administration, 97 Sos. Bucuresti-Ploiesti, Bucharest 013686, Romania f University of Bucharest, Faculty of Geography, Bucharest, Romania
b c

a r t i c l e i n f o
Article history: Received 15 November 2011 Accepted 27 November 2012 Available online Keywords: Clear sky models Global solar radiation Hourly irradiation Romania

a b s t r a c t
Fifty-four broadband clear-sky models for computation of global solar irradiance on horizontal surfaces are tested by using measured data from Romania (South-Eastern Europe). The input data to the models consist of surface meteorological data, column integrated data and data derived from satellite measurements. The testing procedure is performed in twenty-one steps for two different sites in Romania. The models accuracy is reported for various sets of input data. No model ranked the best for all sets of input data. However, some of the models were ranked among the best for most of the testing steps, and thus performed signicantly better than others. These better models are, on an equal footing, ESRA3, Ineichen, METSTAT and REST2 (version 8.1). The next better models are, on an equal footing, Bird, CEM and Paulescu and Schlett. Details about the accuracy of each model are found in the Electronic Supplementary Content for all testing steps. 2012 Elsevier Ltd. All rights reserved.

1. Introduction Although there are several world maps or datasets of solar radiation, they are usually not detailed enough to be used for the determination of available solar energy over small areas. These circumstances have prompted the development of calculation procedures to provide radiation estimates for areas where measurements are not carried out, or for situations when gaps in the measurement records occurred. These procedures range from very simple radiation models to sophisticated computing codes. Only a minority of these models has been validated by their authors, and usually under specic geographical or climatic circumstances only. To increase the condence in modeled data accuracy there is a need for validation by independent groups and at a variety of test sites in many different climatic areas. Existing validation reports usually

* Corresponding author. Candida Oancea Institute, Polytechnic University of Bucharest, Spl. Independentei 313, Bucharest 060042, Romania. Tel.: 40 21 402 9339; fax: 40 21 318 1019. E-mail addresses: badescu@theta.termo.pub.ro (V. Badescu), Chris@ SolarConsultingServices.com (C.A. Gueymard), sorincheval@yahoo.com (S. Cheval). 0960-1481/$ e see front matter 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.renene.2012.11.037

refer to a small number of models being intercompared under specic climatic environments [1e10]. The correct validation and comparison of radiative models raise specic issues. For instance, different statistics may be used to evaluate the bias and random differences between the computed and measured data series. Moreover, various ranking procedures can be used to compare the models accuracy, yielding different results (for more detailed discussions, see [9]). Clear sky solar radiation models are traditionally used in building design and simulation of solar concentration systems. Also, they are of interest for lling historical missing measurement data. Most recent procedures to estimate ground level solar radiation from satellite data are associated with a specic clear sky model. As an example, the Heliosat-2 procedure was developed with special reference to the ESRA clear-sky model [11]. Many other clear sky models are now available and they may be used within other solar radiation computing procedures based on satellite data. The present investigation considers fty-four clear-sky global solar radiation models, i.e., a much larger sample of what the literature offers than in any previous study. This large sample encompasses a large range of modeling complexity, from very simple to sophisticated models. Some models only calculate global

Author's personal copy

86

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

irradiance, whereas others also provide its direct and diffuse components. Most of these models have been previously tested, under a few geographical conditions, during different time intervals and by using various testing procedures. How they compare in practice under specic and identical conditions is still not known. One frequent problem occurring in practice when one has to make use of a radiation model is related to the input data that it requires vs. what is actually available for the specic location and period under scrutiny. It is quite frequent that the potential user of a specic model known to offer good accuracy has access to only databases that do not provide all the necessary input parameters. In this case the user can either obtain the missing data by interpolation, extrapolation or estimation (thus introducing larger uncertainties in the predicted irradiances), or choose another model that would t the available input databases better, but would be presumably simpler and coarser, and thus would have lower accuracy. Compared to the existing validation studies, the present contribution describes some further steps that are deemed necessary to better validate and rank the accuracy of radiation models under the typical conditions encountered by users in practice. In particular, this means that the precise state of the atmosphere cannot be obtained from collocated instruments (such as sunphotometers), whose availability is normally critical in assessing the inherent accuracy of models, as shown in most validation studies mentioned above. The objectives of this investigation are twofold. First, the accuracy of these fty-four models is analyzed and ranked by comparing their predictions under identical climate conditions and using a unique testing procedure. Second, various input datasets are used and the accuracy of each model is reported. Reference radiometric measurements (used as ground truth) are provided here by two meteorological and radiometric stations in South-Eastern Europe (Romania). These stations have provided good-quality routine data over many years. These are not high-end research-class stations, however, which means that the present study can be considered representative of what can be obtained at hundreds of similar stations over the world, rather than at a few specialized sites, like in some previous studies (e.g. [9,10]). These studies considered the ideal case when all the investigated models inputs were measured locally at high frequency with high-quality instruments, so as to obtain the intrinsic performance of these models by avoiding propagation of errors. In contrast, the present investigation is much more pragmatic, since it uses normal radiation measurements and interpolated/extrapolated input data in both space and time to evaluate the real world performance of models under non-ideal conditions.

interested in a particular model will nd here specic information which in [13] is missing. 3. Input data The fty-four solar radiation models have different requirements for input data (Table 2). Table 3 shows the entries needed by each model. The only input variable that is common to all models is the zenith angle, Z, which characterizes the sun position. Most models take the solar geometry into account through the relative air mass, m, rather than Z. The specic relationship between m and Z recommended by each model is used here. Computations were performed only for Z < 85 to avoid inaccuracies resulting from possible horizon shading or experimental cosine errors, for instance. For those models that refer to the solar constant, a common value of 1366.1 W/m2 is used [14]. The input data have been organized in several subgroups. The most important of these consists of measured meteorological data at the ground surface. In Romania, meteorological stations routinely measure the quantities described in Table S1 (prex S come from the Electronic Supplementary Content). 3.1. Solar radiation measurements Two types of measurement devices are used in parallel in each Romanian station, i.e., Robitsch actinographs (which recorded data during a few decades) and Kipp & Zonen CM radiometers (which started recording data in 2006). A CM11 radiometer is operating at Bucharest-Afumati station. All other radiometric stations are provided with CM6B radiometers. The measurement uncertainty is 3% for CM11 and 5% for CM6B. The temperature dependence of sensitivity is 1% for CM11 and 2% for CM6B, on the interval 20 to 40  C. On a monthly basis the bias for CM6B ranges between e 2% and 0.9% [21]. More information about denition of instrument uncertainty may be found in [22,23]. The radiometers are checked twice per week and cleaned when necessary. The measurement methodology is provided by standard procedures prepared at the National Meteorological Administration. Measurements are performed as follows. Solar irradiance (units: W/m2) is measured at 1-min intervals. The series of irradiance values are averaged over 10 min, 60 min and 1440 min, respectively. Irradiation values (units: J) for 10 min, 1 h and 24 h are obtained by multiplying the appropriate average irradiance values by the appropriate time duration. The integration interval starts hour before the time stamp in the les and ends hour after that time stamp. Central European Time is used in the les. The radiometers are calibrated once per year through shaded unshaded measurements in clear sky days and through direct irradiance measurements on a horizontal surface with reference to the Linke-Feussner etalon actinometer. The Linke-Feussner etalon is calibrated with reference to the national etalon, i.e. an Angstrom 702 pyrheliometer with electric compensation. The national etalon is calibrated once at ve years with reference to the World Radiometric Reference at Davos (Swiss). In this study data provided during 2009 by CM K&Z radiometers is used. Partial sky obstructions or shading of the sensors by natural or articial structures is low and is not considered. There are less than 0.1% missing or suspicious data. 3.2. Surface meteorological data Meteorological data measured during 2009 at the three stations of Table 4 are used in this study. In the case of Bucharest, some of the meteorological variables are measured at the Bucharest-

2. Radiation models Fifty-four broadband models for the prediction of global irradiance on horizontal surfaces have been selected for this study. These models are listed in Table 1 using a call number (G001 to G054) for further reference and are briey described the Electronic Supplementary Content (ESC.pdf) where Fortran routines are also provided. Some of these models were already tested in [9,10,12]. Others were intentionally added here for completeness. A birds eye view on all models hierarchy has been reported in [13]. However, no detailed quantitative information about the models accuracy is provided in [13], where the procedure was intended just to check whether a given model passes (or not) some accuracy criteria. The successful events were counted and the counts were used to create hierarchies. In the present paper we provide detailed, quantitative results, for the accuracy of each model in part. Therefore, the reader

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103 Table 1 List of models tested here. More detailed description is provided in the Electronic Supplementary Content (ESC.pdf). Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054 Model name ASHRAE72 ASHRAE05 Badescu Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281 Short description ASHRAE model 1972 ASHRAE model 2005 Badescu model Composite model of Bashahu & Laplaze Model by Barbaro et al. Biga and Rosa model Bird model Atwater & Ball model Chandra model Cloud Layer Sunshine model by Suckling & Hay CPCR2 model Dogniauxs model Daneshyar-Paltridge-Proctor model ESRA model e rst version ESRA model e second version ESRA model e third version ESRA model e fourth version Hottel model Ideriahs model Ineichens model Iqbals Model A Iqbals Model B Iqbals Model C Model by Josefsson Modied Kasten model Kasten model King & Buckius model Model by Krarti McMaster model Model by Machler & Iqbal METSTAT model MRM model version 4 MRM model version 5 Nijegorodov et al. model NRCC model Paltridge & Platt empirical model Perrin de Brichambaut model Psiloglou revised model PSI Model Gueymards REST2 model version 5.0 John Pages team model Composite model described by Carroll Santamouris model Schulze model Sharma & Pal empirical model Watt model Wesely & Lipschutz model Yang model Model by Zhang Combination Hourwith-Schulze models Adnot-Bourges-Campana-Gicquel-Schulze model Model by Paulescu and Schlett Model by Janjai Gueymards REST2 model version 8.1 Table 2 Input data for models used in this work. Symbol Astronomical Year Month Day h Number of year Number of the month in the year (1e12) Number of day in the month (1e31) Hour (UTC) Sun declination (deg.) Zenith angle (deg.) Solar constant (W/m2) Extraterrestrial irradiance (W/m2), corrected for the actual sun-earth distance Kastens air mass Kasten and Youngs air mass Meaning

87

d
z Esc En0 mK mKY Geographical

Latitude (deg) Longitude (deg) Sites elevation (meter) rg Ground albedo Meteorological (surface) p Surface air pressure (hPa) T Air temperature, dry-bulb (K) DT Dry-bulb temperature T at time t, T(t), minus T(t3 hours) U Surface air relative humidity (%) W Wind speed (m/s) Meterological (column integrated) uo Reduced ozone vertical pathlength (atm-cm) uN Total NO2 vertical pathlength (atm-cm) w Precipitable water (cm) Quantities related to atmospheric turbidity a Angstrom wavelength exponent b Angstrom turbidity a1 Angstrom wavelength exponent for <700 nm a2 Angstrom wavelength exponent for >700 nm 61 Aerosol single-scattering albedo, <700 nm 62 Aerosol single-scattering albedo, >700 nm TL1 Linke turbidity estimated by using Page/Remund method [15] TL2 Linke turbidity estimated by using Ineichen [7] method TL3 Linke turbidity estimated from the empirical formula of Dogniaux [16] as a function of precipitable water and Angstroms beta coefcient TL4 Linke turbidity estimated from the average of four linear relationships between Angstroms beta coefcient and Linke turbidity by Hinzpeter [17], Katz et al. [18], Abdelrahman et al. [19] and Grenier et al. [20]. sa Unsworth-Monteith broadband turbidity coefcient s700 Aerosol optical depth at 700 nm, dimensionless Vis Visibility (km)

4 hg

3.3.2. Precipitable water Data for column-integrated precipitable water are available from two stations. Radiosonde measurements are performed twice daily at Bucharest-Baneasa (0000 and 1200 UTC) and once daily at Cluj-Napoca (0000 UTC). Daily measurements for precipitable water performed during 2009 at Cluj-Napoca and BucharestBaneasa are shown in Figs. S2 and S3, respectively.

Baneasa station, whereas other parameters including solar radiation are measured at the Bucharest-Afumati station. 3.3. Column integrated data 3.3.1. Ozone Long-term measurements of column-integrated ozone are performed in Romania once per day at a single station (BucharestBaneasa). Measurements are performed usually in the interval 0900-1400 (Eastern European Time). There are missing days in the recordings (Saturdays, Sundays as well as other days). The WOUDC archive provides data for Bucharest during 1980e2006. The ozone data measured in 2009 are used here, as shown in Fig. S1.

3.4. Satellite derived data 3.4.1. Surface albedo The ground albedo is obtained for all stations from satellite images on a monthly basis and 15-km spatial resolution. We have used the Surface Albedo (SAL) product of the Satellite Applications Facility on Climate Monitoring (CM-SAF) (http://www.cmsaf.eu), which is part of the EUMETSAT distributed ground segment. The product line covers broadband albedo products from the SEVIRI instrument aboard Meteosat-9 [24]. No information about topography or terrain roughness around the weather stations was included in the input data les at this stage. Figs. S4 and S5 show

Author's personal copy

88 Table 3 Input data for tested models. Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054 Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281 z X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X m

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

rg

Vis

uo

uN

sa

TL

X X X X X X X X

X X

X X X X X X X

X X

X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X

X X X X X X X X X X X X X X

X X X X X X

X X X X

X X X

X X X X X X X X

X X X X X X

X X X X X

X X X X X X X X X X X X X X X X X X X X X

X X

X X X X

X X

X X X

X X X X

X X X

X X

X X X X X

X X X X X

X X X X X X

X X

X X X

X X X X X

X X X

X X

X X

X X X

X X X

X X X

X X X

X X

X X X

monthly albedo information for Cluj-Napoca and BucharestBaneasa as observed during 2009. 3.4.2. Atmospheric turbidity data The ngstrm turbidity coefcients a and b, as well as the aerosol single-scattering albedo 6, are obtained for each station from

world datasets that have been published [25]. These are climatological (long-term monthly average) values, whereas daily values would ideally be necessary. These datasets actually provide the aerosol optical depth at 550 nm, sa550, rather than . The latter must therefore be calculated from ngstrms Law:

b sa550 $0:55a

(1)

Table 4 Meteorological stations involved in this study. Geographic code 430613 430608 647334 Station Bucharest-Afumati Bucharest-Baneasa Cluj-Napoca Latitude (deg N) 44.50 44.50 46.78 Longitude (deg E) 26.21 26.13 23.57 Altitude (m) 90 90 417

Data les containing values of a, sa550 and 6 are the primary databases that are used here to evaluate aerosol extinction in those models that take it into account. Data is gridded at 1 1 resolution, and no correction for elevation is considered. Ideally, a much ner spatial resolution (such as 10 10 km) would be necessary for validation purposes, but this is currently not available. Figs. S4 and S5 show the monthly values of a, s550 and 6 at Cluj-Napoca and

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

89

Bucharest-Baneasa, respectively. The monthly data were smoothed to derive pseudo daily data. A more elaborate value of the Angstrom turbidity coefcient is used in computation. A rst mean monthly estimate of the Angb ) is computed by using Eq. (1). Next, the modal strom turbidity (say b value b is calculated from [25]

  b b b 0:83212 3:2104 b
0:75 b 1 5:852 b

(2)

Some models use a special value of that corresponds to an These various inputs used by different models to evaluate the effects of aerosols are shown in Table 3. 3.5. Default values The column-integrated nitrogen dioxide content is not measured. Constant values of 0.0002 atm-cm over rural areas and 0.0005 atm-cm over cities are assumed. The latter value is adopted here for both Bucharest and Cluj-Napoca.

a value xed at 1.3. It is calculated here by using Eq. (1) with a 1.3.

4. Results and discussions The most common bulk performance statistics are the Mean Bias Error (MBE) and the Root Mean Square Error (RMSE) which are dened by, respectively

MBE

n 1 X e nm i 1 i

(3a)

(between 0 and 1 e instantaneous quantity, denoted by C) and the relative sunshine (between 0 and 1, denoted by RSS). The cloud fraction C is computed by using the total cloud cover amount estimated hourly in octas by trained observers at Cluj-Napoca and Bucharest-Baneasa station. The relative sunshine RSS is computed by using sunshine duration measurements. Before 2006 CampbellStokes devices were use to estimate the sunshine duration. After 2006, the sunshine duration is calculated from solar radiation by using the World Meteorological Organization sunshine criterion [26]. This stipulates that the sun is shining if the normal global solar irradiance exceeds 120 W/m2. Sunshine data are recorded hourly. Daily totals are also computed. Measurements are reported in tens of an hour. In the case of Bucharest, sunshine duration is still measured with a Campbell-Stokes recorder at the BucharestAfumati station. Ideally, one expects that the relation C 1RSS is fullled when long term averages are used. However, it is a well-known issue that deviations exist from this relationship. At Bucharest the deviations may be even larger since C and RSS are estimated in two different places located 10.8 km from each other (see Fig. S6). Indeed, during cloudy days the time sequences with shining sun may be quite different at Bucharest-Baneasa and Bucharest-Afumati, respectively. However, this study refers to clear days only. It is likely that the sky is clear simultaneously at both meteorological stations in Bucharest. This assumption is analyzed in Section 4.1, where the cloud fraction and the relative sunshine are used as input. The testing procedure is performed in several stages intended to provide information about the models accuracy for various sets of input data. Separate analyses are performed for Cluj-Napoca and Bucharest. In both cases details are rst provided about the available inputs while the results are reported next. Finally, an overall conclusion about models performance is drawn.

v u n X 1u t1 RMSE e2 m n i1 i

(3b)

4.1. Accuracy analysis A general approach for sensitivity analysis is provided in [27,28]. The choice of the proper sensitivity analysis technique depends on many factors, including (i) the computational cost, (ii) the number of input factors, (iii) the degree of complexity, (iv) the amount of analysts time involved in the sensitivity analysis, and (v) the objective of the analysis (such as factors xing and mapping) [27]. Here we prefer to use a simpler approach based on variance-type indicators. The purpose here is to see the models performance for different sets of realistic input data. The input data sets are realistic in the sense that some categories of real users may have only partial access to very accurate input data (for instance, input data obtained by very accurate measurements). The usual solution in this case is to replace the missing accurate input data with less accurate input data, resulting from less accurate measurements or from various types of estimations. A rst test (so called Stage 1) consisted of checking the procedure. A large number of subsequent tests have been performed for Cluj-Napoca and Bucharest by using data from 2009 only. The most relevant fourty-two testing stages are reported here. Stages 2e22 refer to Cluj-Napoca while Stages 32e52 refer to Bucharest. These stages are organized in pairs, e.g. Stage 2 and Stage 32 are similar, but they refer to Cluj-Napoca and Bucharest, respectively. 4.1.1. Stage 2 (Cluj-Napoca) and Stage 32 (Bucharest) These testing stages make use of the best-quality input data available. Only data for which C 0 and RSS 1 (simultaneously) are used, resulting in a total of 133 hourly records at Cluj-Napoca and 324 hourly recordings at Bucharest.

where n is the number of measured and computed values, denoted mi and ci (i 1,n), respectively, while:

ei hci mi
n 1X m n i1 i

(4)

Also, the mean value of the measured values, m, is dened by:

mh

(5)

Eq. (3a,b) provide dimensionless quantities. In what follows, they are expressed in percent for clarity. A model designed to compute hourly solar irradiation provides goodp performance if the MBE, RMSE and the standard deviation s RMSE2 MBE2 have as low values as possible. The following quantitative recommendations are sometimes used for computation of global irradiation [9]: MBE within 10% and RMSE < 20% indicate good tting between model results and measurements. The following criteria are used in this work to stratify the hourly irradiation models:  good model means jMBEj < 5% and RMSE < 15%;  good enough model means 5% < jMBEj < 10% and 15% < RMSE < 20%;  bad model means jMBEj > 10% and RMSE > 20%. Two parameters are used to estimate the state of the sky, directly and indirectly, respectively. They are the cloud fraction

Author's personal copy

90

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

Daily radiosonde measurements of precipitable water performed at 0000 UTC at Cluj-Napoca and 0.00 and 12.00 UTC at Bucharest-Baneasa are used. For intermediate hours, values are linearly interpolated between two consecutive measured values. Column integrated ozone is not measured at Cluj-Napoca, thus daily measurements performed at Bucharest (once per day at 1200 UTC) are used for both Cluj-Napoca and Bucharest. For intermediate hours, values were linearly interpolated. Monthly-averaged satellite-derived values of ground albedo (rg), a, sa550 and 6 for Cluj-Napoca and Bucharest are used. These values are associated to the 15th day of each month. For intermediate days, values were linearly interpolated. These parameters are assumed constant during any given day. Table 5 shows the accuracy indicators for all models while Fig. 1 shows the results for those models fullling the performance criteria indicated above at Cluj-Napoca. Models G053, G052 and G035 provide the best performance. Fig. 2 shows a comparison between the measured data and the results predicted by model G053. Table S2 provides the input data used by the three models fullling the performance criteria. Table 6 shows the accuracy indicators for all models while Fig. 3 shows the results for the models fullling the performance criteria at Bucharest. Models G053, G003 and G032 have the best performance. Fig. 4 shows a comparison between the measured data and the results predicted by model G053. Table S3 shows the input data used by the models fullling the performance criteria. The accuracy criteria are not fullled by simple models but only by more complex models. 4.1.2. Stage 3 (Cluj-Napoca) and Stage 33 (Bucharest) Here we tested the effect of input data quality related to visual (direct) estimation of the state of the sky. All the data for which C 0 are used (while RSS is a free parameter). Other input data are similar to those used in Stage 2. A total of 166 hourly recordings at Cluj-Napoca (Fig. S7) and 391 hourly recordings at Bucharest (Fig. S8) fullled the test condition. Only a few data associated with both C 0 and RSS 0 exist. Such situations may occur during clear days near sunrise and sunset, i.e., when the irradiance is too low to activate the sunshine recorder. Fig. S7 shows that Stage 3 uses input data associated to RSS lower than unity. Therefore, one expects lower model accuracy in Stage 3 than in Stage 2. However, by comparing Table S4 to Table 5, it appears that models accuracy generally increase at Cluj-Napoca by using the more relaxed input data set of Stage 3. Even though most models perform in Stage 3, a few models show performance degradation (i.e. G009; G018; G035; G042; G044; G052). Table 7 provides an easier comparison between results obtained for Stages 2 and 3 at Cluj-Napoca for some selected models. Generally, both MBE and RMSE decrease slightly in Stage 3 compared to Stage 2 while standard deviations increase for all models. An exception is model G052. A brief explanation follows. The slight decrease of MBE in Stage 3 may be the result of the larger number of data compared to Stage 2. This is the known effect of regression towards the mean [29]. The larger standard deviations obtained at Stage 3 take into account the expected increase in error scattering as compared to Stage 2. The decrease of RMSE is the result of these two opposite effects, in which the inuence of the decreasing MBE values is stronger. Stage 33 below provides further explanations. Table S5 shows that models accuracy apparently increases at Bucharest by using the more relaxed set of input data of Stage 33 (compare with Table 6 for stage 32). Some of the comments associated to Stage 3 for Cluj-Napoca apply here, too. For instance, both the MBE and the standard deviation decrease in Stage 33 for most models as compared to Stage

32. However, the RMSE increases in Stage 33 for most models as compared to Stage 32. This didnt happen when Stage 3 was compared to Stage 2. A few comments follow. The decrease of MBE is the result of the effect of regression towards the mean discussed at Stage 3. The larger standard deviation in Stage 33 takes into account the expected increased error spreading as compared to Stage 32. The apparent increase of RMSE is the result of these two opposite effects, where the inuence of the standard deviation decreasing is stronger. This is different from Stage 3, where the inuence of MBE decreasing is more important. 4.1.3. Stage 4 (Cluj-Napoca) and Stage 34 (Bucharest) Here we tested the effect of input data quality related to the (indirect) estimation of the state of the sky by means of sunshine data. The simple approximation C 1RSS is sometimes used in literature when data on cloud fraction is missing. Only data associated to RSS 1 were used here, while no constraint has been imposed to C. Most of the cloud fraction values equal 0, as expected. However, positive values of C do also exist. This corresponds to periods with the sun shining during the whole hourly interval, while some parts of the sky are covered by clouds. 775 hourly recordings were used at Cluj-Napoca (Fig. S9) and 908 hourly recordings were used at Bucharest (Fig. S10). Other input data are similar to the input data of Stage 2. Table S6 shows that the accuracy of all models signicantly decreases at Cluj-Napoca by using this more relaxed set of input data (compare with Table 5). Consequently, most models do not meet the performance criteria. This proves that using clear-sky models under faint cloudy conditions is not recommended. Also, the models accuracy decreases at Bucharest by using the more relaxed set of input data of Stage 34 as compared to Stage 32 (compare Table S7 and Table 6). 4.1.4. Stage 5 (Cluj-Napoca) and Stage 35 (Bucharest) These testing stages intend to estimate the model accuracy when a constant ozone value uo 0.3 atm-cm is assumed for the whole year. The other input data are those used for Stage 2, providing 133 hourly recordings for Cluj-Napoca and 324 hourly recordings for Bucharest. The models may be stratied as: (i) those not using ozone data as input; and (ii) those using it. Models of the rst category behave similarly in Stages 5 and 2 at Cluj-Napoca, and Stages 35 and 32 at Bucharest, as expected. Table S8 shows that the accuracy of all models of category (ii) degrades at Cluj-Napoca when using a xed amount of ozone (compare with Table 5). Table 8 provides a summary of results for some models of category (ii): MBE increases in comparison with Stage 2, as expected. The RMSE of a few models decreases in Stage 5. This may be explained by a compensation effect: even though inexact ozone data are used, the bias they introduce may compensate for other error sources. The standard deviation is larger in Stage 5 than in Stage 2 for all models, since the decrease of (the absolute values of) MBE is always larger than the decrease of RMSE. Table S9 shows that in case (ii) the models accuracy slightly decreases at Bucharest by using a x amount of ozone as input data (compare with Table 6). There is, however, one model (G052) whose accuracy increases. Table 9 provides an easier comparison for models of category (ii). 4.1.5. Stage 6 (Cluj-Napoca) and Stage 36 (Bucharest) These testing stages intend to estimate the models accuracy when precipitable water is computed by using the formula by Leckner [30]. A total of 133 hourly recordings at Cluj-Napoca and 324 hourly recordings at Bucharest are used. All other inputs are those of Stage 2.

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

91

Table 5 Accuracy indicators at Cluj-Napoca for Stage 2.

Color

Meaning Best model Good model Good enough model Bad model Worst model |MBE|<5%; RMSE<15% 5%<|MBE|<10%;15%<RMSE<20% |MBE|>10%; RMSE>20%

Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE2 (%) 2.850 7.590 2.540 6.190 17.860 2.630 3.980 4.440 -4.150 14.120 7.480 6.350 18.160 8.660 7.120 4.260 4.780 -3.550 39.600 3.420 10.180 6.790 5.210 5.190 5.380 4.860 10.100 -6.880 10.080 4.110 3.890 2.050 4.930 9.200 -1.110 9.710 7.020 5.090 7.840 6.260 8.160 -2.690 -4.930 -2.390 2.420 13.890 -29.070 5.760 -4.710 17.530 5.970 -0.880 0.090 3.890

RMSE2 (%) 6.560 9.550 6.160 8.270 20.130 7.410 6.920 7.160 8.570 15.240 9.310 8.650 20.000 10.510 9.200 6.920 7.910 7.230 41.640 6.490 11.610 8.780 7.700 7.550 7.740 7.820 11.470 13.790 11.540 9.220 6.850 7.300 7.400 10.720 7.580 12.000 9.350 7.510 9.520 8.340 10.040 6.310 7.570 7.160 6.840 15.750 30.880 7.960 12.640 18.790 9.040 5.840 7.420 6.810

Author's personal copy

92

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

the RMSE values decrease in Stage 36 for these models. Thus, the accuracy is generally better in Stage 36. There are a few other models (i.e. G030; G052; G053) with negative MBE values in Stage 36 and the associated decreasing of MBE as reported to Stage 32 means the bias error increases for these models. The RMSE values increase in Stage 36 for some of these models (i.e. G052, G053). Thus, the accuracy is generally worse for these models in Stage 36. For the other models (i.e. G030) the RMSE values decrease in Stage 36. But the decrease of RMSE is smaller than the decrease of MBE and this means the accuracy in Stage 36 is worse for these models, too, as compared to Stage 32. 4.1.6. Stage 7 (Cluj-Napoca) and Stage 37 (Bucharest) These testing stages intend to estimate the models accuracy when the conventional constant value of 0.2 is used for the ground albedo all over the year. The other input data are those of Stage 2. A total of 133 hourly recordings were used at Cluj-Napoca and 324 hourly recordings were used at Bucharest. The models may be stratied as: (i) not using albedo as input data and (ii) using albedo data. For category (i) of models the accuracy at Cluj-Napoca in Stage 7 is similar to the accuracy of Stage 2 while at Bucharest Stage 37 and Stage 32 have the same accuracy. The comparison of Table 5 and Table S12 shows that, in general, using a xed albedo value of 0.2 at Cluj-Napoca results in accuracy degradation for the models of category (ii). Table 12 provides a summary of results for some of these models. For most models both MBE and RMSE slightly increase compared to Stage 2. One may conclude that using an observed surface albedo is recommended. Table S13 shows that generally in case (ii) the models accuracy decreases at Bucharest by using a computed value instead of a measured albedo value (compare with Table 6). Table 13 provides a more detailed comparison for models of category (ii). For all models both MBE and RMSE slightly increase in Stage 37 as compared to Stage 32. Thus, the accuracy is a bit better in Stage 32. This is to be compared with results at Cluj-Napoca where the inuence of the albedo input quality on models accuracy is less conclusive. 4.1.7. Stage 8 (Cluj-Napoca) and Stage 38 (Bucharest) Here we tested the effect of input data quality related to both the visual (direct) estimation of the state of the sky and ozone content in the atmosphere. Only data for C 0 were used (i.e. the RSS values are free to vary). A total of 166 hourly recordings were used for ClujNapoca and 391 hourly recordings for Bucharest. Other input data are those of Stage 5 (i.e. with a constant ozone value used as input). Table S14 shows that for some models the MBE values are larger at Cluj-Napoca in Stage 8 than in Stage 5. The opposite happens for other models. The same comments may be made for the RMSE values. Table S15 shows that for some good models the MBE values are larger in Stage 38 than in Stage 35 at Bucharest. However, the opposite happens for the same models in case of the RMSE values. There are other (worse performing) models whose MBE and RMSE values decrease in Stage 38 as compared to Stage 35. One may conclude that no clear correlation exists between the models accuracy and the combined effect of the quality of ozone input data and clear sky estimations. 4.1.8. Stage 9 (Cluj-Napoca) and Stage 39 (Bucharest) Here we tested the effect of input data quality related to both the visual (direct) estimation of the state of the sky and the precipitable water content in the atmosphere. Only data for C 0 were used (i.e. the RSS values are free to vary). A total of 166 hourly recordings were used for Cluj-Napoca and 391 hourly recordings for Bucharest. Other input data are those of Stage 6 (i.e. with precipitable water computed by using Leckner (1978) formula).

Fig. 1. MBE and standard deviation for the models fullling the performance criteria for computation of global hourly irradiation at Cluj-Napoca in 2009. The model number is attached to the column.

Two categories of models exist: (i) those not using precipitable water as input; and (ii) those using it. The accuracy of all the models of the rst category is the same in Stages 6 and 2 at Cluj-Napoca and Stages 36 and 32 at Bucharest. Comparison of Table S2 and Table S10 shows that at Cluj-Napoca the accuracy of category (ii) models may increase or may decrease. This is also shown in Table 10, which provides a summary. In Stage 6, most models exhibit a positive MBE, although it is less than in Stage 2. This is also the case with RMSE. One may conclude that either the use of calculated precipitable water each hour is better than that of interpolated daily measurements, or that, like in Stage 5, there are compensations of errors. A few models (G035, G042, G052, and G053) underestimate more in Stage 6 than in Stage 2, while their RMSE increase. One may conclude that no compensation of errors exist for these models at this Stage 6. In case (ii) at Bucharest the models accuracy may increase or may decrease by using a computed value instead of a measured value of precipitable water (compare Table S11 with Table 6). Table 11 provides an easier comparison for some models of category (ii). In Stage 6 many models have positive MBE values. The bias error decreases for these models as compared to Stage 32. Also,

Fig. 2. Comparison between results predicted by model G053 and measured data at Cluj-Napoca in 2009.

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

93

Table 6 Accuracy indicators at Bucharest for Stage 32. Color code similar to Table 5.

Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE (%) 2.210 6.730 0.130 3.590 17.370 2.200 2.310 1.150 -5.450 11.320 5.520 3.580 17.540 5.490 4.160 0.450 1.830 -7.410 36.700 0.770 7.490 4.220 3.250 2.900 2.960 2.960 8.020 -6.300 7.680 -1.150 2.000 0.160 1.980 7.390 -2.690 8.810 5.610 2.370 5.160 3.930 5.890 -6.390 -6.510 -2.910 1.870 13.340 -33.060 3.080 -4.570 16.440 5.200 -3.270 -0.130 1.240

RMSE (%) 10.290 12.120 9.680 10.500 21.130 10.340 10.140 10.190 11.910 14.890 11.180 10.950 20.880 11.300 10.700 10.120 10.200 12.480 39.890 10.010 12.290 10.810 10.370 10.100 10.110 10.480 12.760 14.420 12.340 13.350 9.970 10.430 10.180 12.120 11.460 13.440 11.230 10.250 11.000 10.500 11.530 11.780 11.850 10.400 10.170 16.980 35.110 10.420 13.010 19.220 11.310 10.800 11.120 9.940

Performance at ClujNapoca for Stage 2

The worst model

The best model

Author's personal copy

94

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

happens for other models. The same comments may be made for the RMSE values. Table S19 shows that for some good models the MBE values are larger in Stage 40 than in Stage 37 at Bucharest. However, the opposite happens for the other models in case of the RMSE values. There are other (worse performing) models whose MBE and RMSE values decrease in Stage 40 as compared to Stage 37. One may conclude that no clear correlation exists between the models accuracy and the combined effect of the quality of ground albedo input data and clear sky estimations. 4.1.10. Stage 11 (Cluj-Napoca) and Stage 41 (Bucharest) These testing stages intend to estimate models accuracy when measurements about ozone, precipitable water and ground albedo do not exist but the best available surface measurement data are used. A total of 133 hourly recordings were used for Cluj-Napoca and 324 hourly recordings for Bucharest. A constant ozone amount 0.3 atm-cm has been assumed. Precipitable water has been computed by using Leckner formula [30]. A constant ground albedo value (0.2) has been assumed. Other input data are those of Stage 2. The accuracy of some models is unchanged at Cluj-Napoca when passing from Stage 2 to Stage 11 (Table 14). Similarly, the accuracy of the same models is unchanged at Bucharest when passing from Stage 32 to Stage 41 (Table 15). These models do not use as input data values of ozone, precipitable water and ground albedo. Table 14 shows that for most models both the MBE and RMSE decrease at Cluj-Napoca in Stage 11 as compared to stage 2. Table 15 shows that for most models the MBE slightly decreases at Bucharest in Stage 41 as compared to stage 32. 4.1.11. Stage 12 (Cluj-Napoca) and Stage 42 (Bucharest) These testing stages intend to estimate models accuracy when measurements about ozone, precipitable water and ground albedo do not exist but only direct (visual) information about the state of the sky exists. Only data for C 0 were used (i.e. the RSS values are free to vary). A total of 166 hourly recordings were used at ClujNapoca and 391 hourly recordings at Bucharest. Other input data are those of Stage 11. Table S20 shows that for most models both the MBE and RMSE values decrease at Cluj-Napoca in Stage 12 as compared to Stage 11. However, exceptions exist. Thus, the accuracy of the models is generally better in Stage 12. Table S21 shows that for most models the MBE slightly decreases at Bucharest and the RMSE slightly decreases in Stage 42 as compared to Stage 41. This is slightly different from the results obtained at Cluj-Napoca. 4.1.12. Stage 13 (Cluj-Napoca) and Stage 43 (Bucharest) These testing stages are associated to the best quality input data during the extended warm season (i.e. during AprileOctober). Only data associated to C 0 and RSS 1 were used. A total of 129 hourly recordings were used at Cluj-Napoca and 289 hourly recordings at Bucharest. Other input data are those of Stage 2. Table 16 shows that the MBE values increase while the RMSE values slightly decrease in Stage 13 as compared to Stage 2 at ClujNapoca. Thus, the accuracy is generally worse in Stage 13. Table S22 shows that for most models the MBE values decrease while the RMSE values slightly increase in Stage 43 as compared to Stage 42 at Bucharest. This is different from the results obtained at Cluj-Napoca. 4.1.13. Stage 14 (Cluj-Napoca) and Stage 44 (Bucharest) These testing stages are associated to the best quality input data during the extended cold season (i.e. during NovembereMarch). Only data associated to C 0 and RSS 1 were used. A total of 4

Fig. 3. MBE and standard deviation for the models fullling the performance criteria for computation of global hourly irradiation at Bucharest in 2009. The model number is attached to the column.

Table S16 shows that for some good models the MBE values are smaller in Stage 9 than in Stage 6 at Cluj-Napoca. The opposite happens for other models. The same comments may be made for the RMSE values. Table S17 shows that for some good models the MBE values are larger in Stage 39 than in Stage 36 at Bucharest. However, the opposite happens for the same models in case of the RMSE values. There are other (worse performing) models whose MBE and RMSE values decrease in Stage 39 as compared to Stage 36. One may conclude that no clear correlation exists between the models accuracy and the combined effect of the quality of precipitable water input data and clear sky estimations. 4.1.9. Stage 10 (Cluj-Napoca) and Stage 40 (Bucharest) Here we tested the effect of input data quality related to both the visual (direct) estimation of the state of the sky and the ground albedo. Only data for C 0 were used (i.e. the values of the sunshine fraction are free to vary). A total of 166 hourly recordings were used for Cluj-Napoca and 391 hourly recordings for Bucharest. Other input data are those of Stage 7 (i.e. with xed value of ground albedo). Table S18 shows that for some good models the MBE values are smaller in Stage 10 than in Stage 7 at Cluj-Napoca. The opposite

Fig. 4. Comparison between results predicted by model G053 and measured data at Bucharest in 2009.

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

95

Table 7 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 3, respectively; DMBE=RMSE=shMBE=RMSE=s3 MBE=RMSE=s2 . Color code similar to Table 5.

hourly recordings were used at Cluj-Napoca and 289 hourly recordings at Bucharest. Other input data are those of Stage 2. Table 17 shows a great variety of results for Cluj-Napoca. For some models the MBE values increase in absolute value during the extended cold season. For other models the reverse situation happens. Also, the RMSE values increase or decrease as compared to Stage 2. Note, however, that these results should be taken with caution since only 4 recordings are associated to the extended cold season. Table S23 shows a great variety of results for Bucharest. They are basically similar to those shown in Table 17. 4.1.14. Stage 15 (Cluj-Napoca) and Stage 45 (Bucharest) These testing stages intend to estimate models accuracy during the extended warm season (i.e. during AprileOctober) as a function of input data related to visual (direct) estimation of the state of the sky. Only data associated to C 0 were used while the values of RSS are free to vary. A total of 161 hourly recordings were used for ClujNapoca and 328 hourly recordings for Bucharest. Other input data are those of Stage 3. Table S24 shows that for some models the MBE values slightly increase while the RMSE values slightly decrease in Stage 15 as compared to Stage 3 at Cluj-Napoca. Other models are characterized by a reverse statistical indicators variation. Table S25 shows that the MBE values slightly increase or decrease in Stage 45 as compared to Stage 33 at Bucharest. However, the RMSE values decrease in Stage 45 for most models. 4.1.15. Stage 16 (Cluj-Napoca) and Stage 46 (Bucharest) These testing stages intend to estimate models accuracy during the extended cold season (i.e. during NovembereMarch) as a function of input data related to visual (direct) estimation of the state of the sky. Only data associated to C 0 were used while the values of RSS are free to vary. A total of 5 hourly recordings were used for Cluj-Napoca and 63 hourly recordings for Bucharest. Other input data are those of Stage 3.

Table S26 shows a great variety of results for Cluj-Napoca. For some models the MBE values increase in absolute value during the extended cold season. For other models the reverse situation happens. Also, the RMSE values increase or decrease as compared to Stage 2. Note, however, that these results should be taken with caution since only 5 recordings are associated to the extended cold season. Table S27 shows a great variety of results for Bucharest. For most models the MBE values increase in absolute value during the extended cold season. For a few other models the reverse situation happens. Also, the RMSE values increase or decrease as compared to Stage 33. This is similar to the results obtained at Stage 16. 4.1.16. Stage 17 (Cluj-Napoca) and Stage 47 (Bucharest) These testing stages intend to estimate models accuracy during the extended warm season (i.e. during AprileOctober) as a function of input data related to (indirect) estimation of the state of the sky by means of sunshine recordings. Only data associated to RSS 1 were used in this stage while the values of C are free to vary. A total of 695 hourly recordings were used for Cluj-Napoca and 874 hourly recordings for Bucharest. Other input data are those of Stage 4. Table S28 shows very little difference between Stage 17 and Stage 4 at Cluj-Napoca. All models have a bad performance in Stage 17. The accuracy slightly increases for those models underestimating global irradiation and slightly decreases for the other models. Table S29 shows very little difference between Stage 47 and Stage 34 at Bucharest. All models have a bad performance in Stage 47. The accuracy generally slightly decreases for most models and slightly increases for a few models. 4.1.17. Stage 18 (Cluj-Napoca) and Stage 48 (Bucharest) These testing stages intend to estimate models accuracy during the extended cold season (i.e. during NovembereMarch) as a function of input data related to (indirect) estimation of the state of the

Table 8 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 5, respectively. Only models using ozone as input data are shown. DMBE=RMSEhMBE=RMSE5 MBE=RMSE2 . Color code similar to Table 5.

Author's personal copy

96

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

Table 9 Comparison between results obtained at Bucharest for Stage 32 and DMBE=RMSEhMBE=RMSE35 MBE=RMSE32 . Color code similar to Table 5.

Stage

35,

respectively.

Only

models

using

ozone

as

input

are

shown.

Model number G003 G007 G022 G023 G031 G032 G033 G038 G040 G048 G052 G053 G054

Model name Badesc Bird IqbalB IqbalC METSTAT MRM4 MRM5 PR REST250 Yang Paulescu Janjai REST281

MBE35 (%) 0.24 2.43 4.35 3.38 2.13 0.22 2.11 2.51 4.05 3.19 -3.17 0.92 1.35

MBE (%)
0.11 0.12 0.13 0.13 0.13 0.06 0.13 0.14 0.12 0.11 0.1 1.05 0.11

RMSE35 (%) 9.67 10.15 10.85 10.39 9.98 10.42 10.2 10.28 10.53 10.45 10.76 10.67 9.96

RMSE
(%) -0.01 0.01 0.04 0.02 0.01 -0.01 0.02 0.03 0.03 0.03 -0.04 -0.45 0.02

sky by means of sunshine recordings. Only data associated to RSS 1 were used in this stage while the values of C are free to vary. A total of 65 hourly recordings were used for Cluj-Napoca and 95 hourly recordings for Bucharest. Other input data are those of Stage 4. Table S30 shows that generally the accuracy increases in Stage 18 as compared to Stage 4 at Cluj-Napoca. In some other cases the accuracy decreases in Stage 18. The increase or decrease of accuracy between the two Stages is a function of model. However, all models have poor performance in Stage 18. Table S31 shows that the accuracy generally strongly decreases at Bucharest in Stage 48 compare to Stage 34. In a few cases the reverse situation happens. However, the models have generally poor performance in Stage 48. 4.1.18. Stage 19 (Cluj-Napoca) and Stage 49 (Bucharest) These testing stages intend to estimate models accuracy when measurements about ozone, precipitable water and ground albedo do not exist but the best quality surface measurement data are used during the extended warm season (i.e. AprileOctober). A total of

129 hourly recordings were used in Cluj-Napoca and 289 hourly recordings at Bucharest. Other input data are those of Stage 11. Table S32 shows very little difference between Stage 19 and Stage 11 at Cluj-Napoca. The accuracy generally slightly decreases in Stage 19. In a few cases the reverse situation happens. Table S33 shows that generally the accuracy increases in Stage 49 as compared to Stage 41 at Bucharest. In a few cases the reverse situation happens. 4.1.19. Stage 20 (Cluj-Napoca) and Stage 50 (Bucharest) These testing stages intend to estimate models accuracy when measurements about ozone, precipitable water and ground albedo do not exist but the best quality surface measurement data are used during the extended cold season (i.e. NovembereMarch). A total of 4 hourly recordings were used at Cluj-Napoca and 35 hourly recordings at Bucharest. Other input data are those of Stage 11. Table S34 shows that the accuracy of some models signicantly increases in Stage 20 as compared to Stage 11 at Cluj-Napoca. However, the reversed situation happens for other models.

Table 10 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 6, respectively. Only models using precipitable water as input data are shown. DMBE=RMSEhMBE=RMSE6 MBE=RMSE2 . Color code similar to Table 5.

Model number G003 G007 G008 G020 G030 G031 G032 G033 G035 G042 G052 G053 G054

Model name Badesc Bird CEM Ineich Machlr METSTAT MRM4 MRM5 NRCC RSC Paulescu Janjai REST281

MBE6 (%) 1.85 3.5 3.87 2.6 4.07 3.37 1.76 4.25 -1.61 -3.57 -1.38 -0.1 3.12

MBE (%)
-0.69 -0.48 -0.57 -0.82 -0.04 -0.52 -0.29 -0.68 -0.5 -0.88 -0.5 -0.19 -0.77

RMSE6 (%) 6.03 6.78 6.93 6.22 9.21 6.69 7.28 7.09 7.8 6.79 6.09 7.45 6.52

RMSE

(%)
-0.13 -0.14 -0.23 -0.27 -0.01 -0.16 -0.02 -0.31 0.22 0.48 0.25 0.03 -0.29

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

97

Table 11 Comparison between results obtained at Bucharest for Stage 32 and Stage 36, respectively. Only models using precipitable water as input data are shown. DMBE=RMSEhMBE=RMSE35 MBE=RMSE32 . Color code similar to Table 5.

Model number G003 G007 G008 G020 G022 G023 G024 G025 G030 G031 G032 G033 G035 G038 G040 G048 G052 G053 G054

Model name Badesc Bird CEM Ineich IqbalB IqbalC Josefs KASM Machlr METSTAT MRM4 MRM5 NRCC PR REST250 Yang Paulescu Janjai REST281

MBE36 (%) -0.51 1.86 0.6 -0.03 3.38 2.81 2.16 2.3 -3.16 -1.19 1.53 -0.1 1.35 1.74 3.18 2.44 -3.74 -0.36 0.5

MBE (%)
-0.64 -0.45 -0.55 -0.8 -0.84 -0.44 -0.74 -0.66 -2.01 -3.19 1.37 -2.08 4.04 -0.63 -0.75 -0.64 -0.47 -0.23 -0.74

RMSE36 (%) 9.83 10.16 10.26 10.14 10.67 10.35 10.05 10.06 11.68 13.36 10 10.47 10.2 10.25 10.38 10.37 11.06 11.17 10.02

RMSE
(%) 0.15 0.02 0.07 0.13 -0.14 -0.02 -0.05 -0.05 -1.67 3.39 -0.43 0.29 -1.26 0 -0.12 -0.05 0.26 0.05 0.08

Table 12 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 7, respectively. Only models using ground albedo as input data are shown. DMBE=RMSEhMBE=RMSE7 MBE=RMSE2 . Color code similar to Table 5.

Model number G007 G008 G031 G042 G043 G054

Model name Bird CEM METSTAT RSC Santam REST281

MBE7 (%) 4.12 4.54 4.02 -2.53 -4.81 4.03

MBE (%)
0.14 0.1 0.13 0.16 0.12 0.14

RMSE7 (%) 6.99 7.22 6.92 6.25 7.49 6.89

RMSE

(%)
0.07 0.06 0.07 -0.06 -0.08 0.08

Table 13 Comparison between results obtained at Bucharest for Stage 32 and Stage 37, respectively. Only models using ground albedo as input data are shown. DMBE=RMSEhMBE=RMSE37 MBE=RMSE32 . Color code similar to Table 5.

Model number G007 G008 G022 G023 G024 G031 G033 G038 G040 G054

Model name Bird CEM Iqbal B Iqbal C Josefs METSTAT MRM5 PR REST250 REST281

MBE37 (%) 2.53 1.31 4.3 3.48 3.09 2.21 2.2 2.58 4.17 1.47

MBE (%) 0.22 0.16 0.08 0.23 0.19 0.21 0.22 0.21 0.24 0.23

RMSE37 (%) 10.19 10.21 10.84 10.44 10.15 10.02 10.22 10.3 10.59 9.97

RMSE (%) 0.05 0.02 0.03 0.07 0.05 0.05 0.04 0.05 0.09 0.03

Author's personal copy

98

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

Table 14 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 11, respectively. DMBE=RMSEhMBE=RMSE11 MBE=RMSE2 . Color code similar to Table 5.

Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE11 (%) 2.85 7.59 2.03 5.53 17.69 2.63 3.83 3.97 -4.15 13.44 7.18 5.96 18.16 8.38 6.77 3.68 4.78 -3.55 38.94 2.6 9.64 6.15 5.06 4.5 4.67 4.86 9.65 -6.88 9.46 4.07 3.69 1.86 4.59 8.89 -1.61 9.71 6.79 4.75 7.26 5.8 7.86 -3.41 -5.24 -2.39 2.42 14.44 -29.07 5.23 -4.71 17.53 5.97 -1.22 1.58 3.43

MBE (%) 0 0 -0.51 -0.66 -0.17 0 -0.15 -0.47 0 -0.68 -0.3 -0.39 0 -0.28 -0.35 -0.58 0 0 -0.66 -0.82 -0.54 -0.64 -0.15 -0.69 -0.71 0 -0.45 0 -0.62 -0.04 -0.2 -0.19 -0.34 -0.31 -0.5 0 -0.23 -0.34 -0.58 -0.46 -0.3 -0.72 -0.31 0 0 0.55 0 -0.53 0 0 0 -0.34 1.49 -0.46

RMSE11 (%) 6.56 9.55 6.07 7.9 20.35 7.41 6.91 6.98 8.57 14.64 9.14 8.41 20 10.31 9 6.7 7.91 7.23 41 6.22 11.19 8.37 7.66 7.18 7.35 7.82 11.15 13.79 11.05 9.21 6.82 7.29 7.26 10.5 7.8 12 9.25 7.36 9.13 8.08 9.84 6.71 7.87 7.16 6.84 16.2 30.88 7.69 12.64 18.79 9.04 6.01 6.7 6.65

RMSE

(%)
0 0 -0.09 -0.37 0.22 0 -0.01 -0.18 0 -0.6 -0.17 -0.24 0 -0.2 -0.2 -0.22 0 0 -0.64 -0.27 -0.42 -0.41 -0.04 -0.37 -0.39 0 -0.32 0 -0.49 -0.01 -0.03 -0.01 -0.14 -0.22 0.22 0 -0.1 -0.15 -0.39 -0.26 -0.2 0.4 0.3 0 0 0.45 0 -0.27 0 0 0 0.17 -0.72 -0.16

Author's personal copy

Table 15 Comparison between results obtained at Bucharest for Stage 32 and Stage 41, respectively. Only models using ground albedo as input data are shown. DMBE=RMSEhMBE=RMSE41 MBE=RMSE32 . Color code similar to Table 5.

Model number
G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name
ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE41 (%)
2.21 6.73 -0.4 3.01 18.78 2.2 2.21 0.77 -5.45 10.69 5.29 3.23 17.54 5.2 3.83 -0.04 1.83 -7.41 36.12 -0.03 7.08 3.58 3.16 2.35 2.3 2.96 7.79 -6.3 7.18 -1.19 1.85 -0.04 1.7 7.05 -3.16 8.81 5.38 2.08 4.71 3.53 5.58 -7.1 -6.78 -2.91 1.87 14.04 -33.06 2.56 -4.57 16.44 5.2 -3.63 0.69 0.83

MBE ( %)
0 0 -0.53 -0.58 1.41 0 -0.1 -0.38 0 -0.63 -0.23 -0.35 0 -0.29 -0.33 -0.49 0 0 -0.58 -0.8 -0.41 -0.64 -0.09 -0.55 -0.66 0 -0.23 0 -0.5 -0.04 -0.15 -0.2 -0.28 -0.34 -0.47 0 -0.23 -0.29 -0.45 -0.4 -0.31 -0.71 -0.27 0 0 0.7 0 -0.52 0 0 0 -0.36 0.82 -0.41

RMSE41 (%)
10.29 12.12 9.81 10.43 23.15 10.34 10.21 10.27 11.91 14.52 11.16 10.91 20.88 11.23 10.65 10.23 10.2 12.48 39.36 10.14 12.15 10.72 10.43 10.09 10.06 10.48 12.7 14.42 12.15 13.36 10.03 10.46 10.24 11.98 11.68 13.44 11.17 10.3 10.91 10.49 11.45 12.33 12.09 10.4 10.17 17.49 35.11 10.39 13.01 19.22 11.31 11.01 10.69 10.04

RMSE (%)
0 0 0.13 -0.07 2.02 0 0.07 0.08 0 -0.37 -0.02 -0.04 0 -0.07 -0.05 0.11 0 0 -0.53 0.13 -0.14 -0.09 0.06 -0.01 -0.05 0 -0.06 0 -0.19 0.01 0.06 0.03 0.06 -0.14 0.22 0 -0.06 0.05 -0.09 -0.01 -0.08 0.55 0.24 0 0 0.51 0 -0.03 0 0 0 0.21 -0.43 0.1

Author's personal copy

100

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

Table 16 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 13, respectively. DMBE=RMSEhMBE=RMSE13 MBE=RMSE2 . Color code similar to Table 5.

Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE13 (%) 2.95 7.62 2.73 6.28 18.06 2.88 4.03 4.49 -3.99 14.17 7.54 6.38 18.57 8.83 7.25 4.33 5 -3.27 39.91 3.45 10.23 6.76 5.26 5.28 5.52 4.99 10.07 -6.56 10.16 4.65 4.03 2.47 4.99 9.3 -0.94 10.07 7.19 5.15 7.96 6.31 8.22 -2.45 -4.76 -2.02 2.65 14.07 -28.41 5.82 -4.35 17.65 6.11 -0.8 -0.24 3.98

MBE (%) 0.1 0.03 0.19 0.09 0.2 0.25 0.05 0.05 0.16 0.05 0.06 0.03 0.41 0.17 0.13 0.07 0.22 0.28 0.31 0.03 0.05 -0.03 0.05 0.09 0.14 0.13 -0.03 0.32 0.08 0.54 0.14 0.42 0.06 0.1 0.17 0.36 0.17 0.06 0.12 0.05 0.06 0.24 0.17 0.37 0.23 0.18 0.66 0.06 0.36 0.12 0.14 0.08 -0.33 0.09

RMSE13 (%) 6.55 9.53 6.11 8.27 20.15 7.34 6.92 7.15 8.48 15.22 9.31 8.64 20.03 10.52 9.21 6.92 7.88 7 41.63 6.49 11.59 8.74 7.7 7.55 7.74 7.82 11.42 13.56 11.54 8.67 6.84 7.01 7.4 10.72 7.51 11.98 9.35 7.51 9.53 8.34 10.03 6.11 7.44 6.82 6.77 15.77 30.23 7.96 12.36 18.79 9.04 5.81 7.21 6.81

RMSE (%) -0.01 -0.02 -0.05 0 0.02 -0.07 0 -0.01 -0.09 -0.02 0 -0.01 0.03 0.01 0.01 0 -0.03 -0.23 -0.01 0 -0.02 -0.04 0 0 0 0 -0.05 -0.23 0 -0.55 -0.01 -0.29 0 0 -0.07 -0.02 0 0 0.01 0 -0.01 -0.2 -0.13 -0.34 -0.07 0.02 -0.65 0 -0.28 0 0 -0.03 -0.21 0

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

101

Table 17 Comparison between results obtained at Cluj-Napoca for Stage 2 and Stage 14, respectively. DMBE=RMSEhMBE=RMSE14 MBE=RMSE2 . Color code similar to Table 5.

Model number G001 G002 G003 G004 G005 G006 G007 G008 G009 G010 G011 G012 G013 G014 G015 G016 G017 G018 G019 G020 G021 G022 G023 G024 G025 G026 G027 G028 G029 G030 G031 G032 G033 G034 G035 G036 G037 G038 G039 G040 G041 G042 G043 G044 G045 G046 G047 G048 G049 G050 G051 G052 G053 G054

Model name ASHRAE72 ASHRAE05 Badesc Basha BCLSM Biga Bird CEM Chandr CLS CPCR2 Dognio DPPLT ESRA1 ESRA2 ESRA3 ESRA4 HLJ Ideria Ineich IqbalA IqbalB IqbalC Josefs KASM Kasten King KZHW MAC Machlr METSTAT MRM4 MRM5 Nijego NRCC Paltri Perrin PR PSIM REST250 Rodger RSC Santam Schulz Sharma Watt WKB Yang Zhang HS ABCGS Paulescu Janjai REST281

MBE14 (%) -2.48 6.06 -7.02 1.52 7.67 -10.02 1.83 1.75 -12.38 11.64 4.59 4.95 -2.37 0.2 0.55 0.74 -6.73 -17.82 23.77 1.69 7.76 8.33 2.85 0.72 -1.51 -1.92 11.34 -22.77 6.2 -23.58 -3.01 -19.68 1.91 3.66 -9.72 -8.37 -1.6 2.4 2.06 3.78 4.7 -15.1 -13.6 -21.15 -9.61 4.89 -62.54 2.72 -23.43 11.35 -0.92 -4.67 17.02 -0.57

MBE (%) -5.33 -1.53 -9.56 -4.67 -10.19 -12.65 -2.15 -2.69 -8.23 -2.48 -2.89 -1.4 -20.53 -8.46 -6.57 -3.52 -11.51 -14.27 -15.83 -1.73 -2.42 1.54 -2.36 -4.47 -6.89 -6.78 1.24 -15.89 -3.88 -27.69 -6.9 -21.73 -3.02 -5.54 -8.61 -18.08 -8.62 -2.69 -5.78 -2.48 -3.46 -12.41 -8.67 -18.76 -12.03 -9 -33.47 -3.04 -18.72 -6.18 -6.89 -3.79 16.93 -4.46

RMSE14 (%) 4.7 7.25 8.08 4.14 8.55 10.87 4.28 4.31 13.05 12.25 5.97 6.43 4.57 4.12 4.1 4.14 7.97 18.3 24 4.26 8.67 9.2 4.85 3.93 4.24 4.61 11.93 25.29 7.31 30.76 4.91 20.12 4.33 5.08 10.6 9.28 4.26 4.52 4.31 5.39 6.18 15.63 14.17 21.56 10.46 6.35 63.29 4.71 26.44 12.04 4.32 6.13 17.48 3.88

RMSE (%) -1.86 -2.3 1.92 -4.13 -11.58 3.46 -2.64 -2.85 4.48 -2.99 -3.34 -2.22 -15.43 -6.39 -5.1 -2.78 0.06 11.07 -17.64 -2.23 -2.94 0.42 -2.85 -3.62 -3.5 -3.21 0.46 11.5 -4.23 21.54 -1.94 12.82 -3.07 -5.64 3.02 -2.72 -5.09 -2.99 -5.21 -2.95 -3.86 9.32 6.6 14.4 3.62 -9.4 32.41 -3.25 13.8 -6.75 -4.72 0.29 10.06 -2.93

Author's personal copy

102

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

Table S35 shows that for some models the MBE values increase signicantly at Bucharest in Stage 50 as compared to Stage 41. In a few other cases the reverse situation happens. The RMSE values generally increase in Stage 50. 4.1.20. Stage 21 (Cluj-Napoca) and Stage 51 (Bucharest) These testing stages intend to estimate models accuracy during the extended warm season (i.e. AprileOctober) when measurements about ozone, precipitable water and ground albedo do not exist but only direct (visual) information about the state of the sky exists. Only data for C 0 were used (i.e. the RSS values are free to vary). A total of 161 hourly recordings were used at Cluj-Napoca and 328 hourly recordings at Bucharest. Other input data are those of Stage 12. Table S36 shows a comparable models accuracy at Cluj-Napoca in Stage 21 as compared to Stage 12. The RMSE values generally decrease in Stage 21. Table S37 shows a comparable models accuracy at Bucharest in Stage 51 as compared to Stage 42. Generally, the RMSE values slightly decrease in Stage 50. 4.1.21. Stage 22 (Cluj-Napoca) and Stage 52 (Bucharest) These testing stages intend to estimate models accuracy during the extended cold season (i.e. NovembereMarch) when measurements about ozone, precipitable water and ground albedo do not exist but only direct (visual) information about the state of the sky exists. Only data for C 0 were used (i.e. the RSS values are free to vary). A total of 5 hourly recordings were used at Cluj-Napoca and 63 hourly recordings at Bucharest. Other input data are those of Stage 12. Table S38 shows that for some models the difference in models accuracy in Stage 12 and Stage 22 is large at Cluj-Napoca. In some cases the MBE and/or RMSE values increase in Stage 22. The reversed situation also happens. These results should be considered with caution since only 5 recordings are available for calculation. Table S39 shows that for some models the difference in models accuracy in Stage 52 and Stage 42 is large at Bucharest. In some cases the MBE values and/or RMSE values increase in Stage 22. The reversed situation also happens. 4.2. Overview An overview for results obtained in both localities is useful. We have counted the number of testing Stages per model with accuracy criteria fullled for both MBE and RMSE indicators. This way of classication stratied the models in several accuracy groups and the rst twelve best groups are shown on Fig. 5. Sixteen models belong to the rst three best groups. This number is smaller than the similar number for Cluj-Napoca and Bucharest, respectively. Indeed, the best three groups for each of the two locations refers to 18, 17 and 16 testing Stages passed by a given model, respectively, while in Fig. 5 the best three groups refer to 16, 15 and 14 testing Stages passed by a given model, respectively. Three models belonging to the rst three best groups in Fig. 5 (i.e. G001, G006 and G045) are very simple models (i.e. they do not require meteorological data as input). However, all the four models of the best group (i.e. G016, G020, G031 and G054) are based on rather complex algorithms. Two models belonging to the rst best group (i.e. G020 and G054) and one model belonging to the second best group (i.e. G007) were found among the ve best models as a result of the more accurate analysis described in [10]. Note that the present analysis, based on variance type indicators, does not take into account the non linearity of some models and the interactions between the inputs on the output. Tools proposed in the

Fig. 5. Number of testing stages per model with accuracy criteria fullled for both MBE and RMSE indicators. Global solar radiation in 2009 at Cluj-Napoca and Bucharest is considered. The maximum number of testing stages is 21.

literature are able to provide a more general and systematic approach [27,28]. They can be used to evaluate the sensitivity of models and to provide robust and understandable results. However, such sophisticated sensitivity analyses require lengthy procedures and large amounts of computing resources. They should be preceded by simpler analyses, like the present one, which allow obtaining a list of selected models, candidate for further analysis. 5. Conclusions Fifty-four broadband models for computation of global irradiance on horizontal surface were selected for testing. These models were tested by using input data from two meteorological stations from Romania (South-Eastern Europe). The input data consist of surface meteorological data, column integrated data and data derived from satellite measurements. The testing procedure is performed in fourty-two stages intended to provide information about the accuracy of the models for various sets of input data. There is no model to be ranked the best for all sets of input data. However, some of the models performed better than others, in the sense that they were ranked among the best for most of the testing stages. Fig. 5 shows that these models can be grouped in three categories. The best models are G016, G020, G31 and G54. The second best models are G007, G008 and G052. The third best models are G001, G003, G006, G026, G032, G033, G044, G045 and G053. Models accuracy depends on the quality of input data and site of measurements. Details about the accuracy of each model can be found in the paper and in the Electronic Supplementary Content. Acknowledgments This work was supported in part by a grant of the Romanian National Authority for Scientic Research, CNCS e UEFISCDI, project number PN-II-ID-PCE-2011-3-0089 and by the European Cooperation in Science and Technology project COST ES1002. The authors thank the reviewers for useful comments and suggestions.

Author's personal copy

V. Badescu et al. / Renewable Energy 55 (2013) 85e103

103

Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.renene.2012.11.037. References
[1] Davies JA, McKay DC, Luciani G, Abdel-Wahab M. Validation of models for estimating solar radiation on horizontal surfaces. Downsview, Ontario: Atmospheric Environment Service; 1988. IEA Task IX Final Report. [2] Davies JA, McKay DC. Evaluation of selected models for estimating solar radiation on horizontal surfaces. Solar Energy 1989;43:153e68. [3] Gueymard C. Critical analysis and performance assessment of clear sky solar irradiance models using theoretical and measured data. Solar Energy 1993;51: 121e38. [4] Badescu V. Verication of some very simple clear and cloudy sky models to evaluate global solar irradiance. Solar Energy 1997;61:251e64. [5] Batlles FJ, Olmo FJ, Tovar J, Alados-Arboledas L. Comparison of cloudless sky parameterizations of solar irradiance at various Spanish midlatitude locations. Theor Appl Climatol 2000;66:81e93. [6] Iziomon MG, Mayer H. Assessment of some global solar radiation parameterizations. J Atmos Solar Terr Phys 2002;64:1631e43. [7] Ineichen P. Comparison of eight clear sky broadband models against 16 independent data banks. Solar Energy 2006;80:468e78. [8] Younes S, Muneer T. Clear-sky classication procedures and models using a world-wide data-base. Appl Energy 2007;84:623e45. [9] Gueymard CA, Myers DR. Validation and ranking methodologies for solar radiation models. In: Badescu V, editor. Modeling solar radiation at the earth surface. Berlin: Springer; 2008. p. 479e509. chap. 20. [10] Gueymard CA. Clear-sky solar irradiance predictions for large-scale applications using 18 radiative models: improved validation methodology and detailed performance analysis. Solar Energy 2012. http://dx.doi.org/10.1016/ j.solener.2011.11.011. [11] Rigollier C, Bauer O, Wald L. On the clear sky model of the ESRAdEuropean solar radiation atlasdwith respect to the Heliosat method. Solar Energy 2000; 68:33e48. [12] Badescu V, Cheval S, Gueymard C, Oprea C, Baciu M, Dumitrescu A, et al. Testing 52 models of clear sky solar irradiance computation under the climate of Romania, The workshop solar energy at urban scale. Compiegne, France: Universite de Technologie; May 25-26, 2010. p. 12e5. [13] Badescu V, Gueymard CA, Cheval S, Oprea C, Baciu M, Dumitrescu A, et al. Computing global and diffuse solar hourly irradiation on clear sky. Review and testing of 54 models. Sustainable Energy Rev 2012;16:1636e56. [14] Gueymard CA, Kambezidis HD. Solar spectral radiation. In: Muneer T, editor. Solar radiation and daylight models. New York: Elsevier; 2004.

[15] Remund J, Wald L, Lefevre M, Ranchin T, Page J. Worldwide Linke turbidity information. In: Proc. ISES conf. Stockholm, Sweden: International Solar Energy Society; 2003. [16] Dogniaux R. The estimation of atmospheric turbidity. Proc. advances in European solar radiation climatology. London, UK, U.K: Int. Sol. Energy Soc; 1986. p. 3.1e3.4. [17] Hinzpeter H. Uber Trubungsbestimmungen in Potsdam in dem Jahren 1946 und 1947m Meteor, vol. 4; 1950. p. 1. [18] Katz M, Baille A, Mermier M. Atmospheric turbidity in a semi-rural site. Part I: evaluation and comparison of different atmospheric turbidity coefcients. Solar Energy 1982;28:323e7. [19] Abdelrahman MA, Said SAM, Shuaib AN. Comparison between atmospheric turbidity coefcients of desert and temperate climates. Solar Energy 1988;40: 219e25. [20] Grenier JC, de la Casiniere A, Cabot T. A spectral model of linkes turbidity factor and its experimental implications. Solar Energy 1994;52:303e14. [21] Myers DR, Wilcox SM. Relative accuracy of 1-Minute and daily total solar radiation data for 12 global and 4 direct beam solar radiometers, American solar energy society annual conference buffalo. New York; May 11e16, 2009. [22] Gueymard CA, Myers DR. Evaluation of conventional and high-performance routine solar radiation measurements for improved solar resource, climatological trends, and radiative modeling. Solar Energy 2009;83:171e85. [23] Myers DR. Terrestrial solar spectral distributions derived from broadband hourly solar radiation data. In: Tsai BK, editor. Optical modeling and measurements for solar energy systems III: proceedings of SPIE conference, 2e4 august 2009, San Diego, California. Proceedings of SPIE e the international society for optical engineering, paper No. 74100A, Bellingham, WA. SPIE e The International Society for Optical Engineering; 2009. p. 11. [24] Schulz J, Albert P, Behr H-D, Caprion D, Deneke H, Dewitte S, et al. Operational climate monitoring from space: the EUMETSAT satellite application facility on climate monitoring (CM-SAF). Atmos Chem Phys 2009;9:1687e709, http:// www.atmos-chem-phys.org/9/1687/2009/acp-9-1687-2009.pdf. [25] Gueymard CA, Thevenard D. Monthly average clear-sky broadband irradiance database for worldwide solar heat gain and building cooling load calculations. Solar Energy 2009;83(11):1998e2018. [26] WMO. Guide to meteorological instruments and methods of observation. WMO No. 8. Ch. 8 e Measurement of sunshine duration, pp. I.8.1e1.8.11, http://www.wmo.int/pages/prog/www/IMOP/publications/CIMO-Guide/ CIMO_Guide-7th_Edition-2008.html; 2008. [27] Saltelli A, Ratto M, Tarantola S, Campolongo F. Sensitivity analysis for chemical models. Chem Rev 2005;105(7):2811e28. [28] Saltelli A, Ratto M, Andres T, Campolongo F, Cariboni J, Gatelli D, et al. Global sensitivity analysis e the primer. Newy York: John Wiley & Sons; 2008. [29] Stigler SM. Regression toward the mean, historically considered. Stat Methods Med Res 1997;6(2):103e14. [30] Leckner B. The spectral distribution of solar radiation at the Earths surface e elements of a model. Solar Energy 1978;20:143e50.

Das könnte Ihnen auch gefallen