Available online at www.sciencedirect.com

www.elsevier.com/locate/solener

ScienceDirect

Solar Energy 132 (2016) 25–37

Preliminary survey on site-adaptation techniques forsatellite-derived and reanalysis solar radiation datasets

J. Polo a,⇑, S. Wilbert b, J.A. Ruiz-Arias c, R. Meyer d, C. Gueymard e, M. Suri f, L. Martın g,T. Mieslinger d, P. Blanc h, I. Grant i, J. Boland j, P. Ineichen k, J. Remund l, R. Escobar m,

A. Troccoli n, M. Sengupta o, K.P. Nielsen p, D. Renne q, N. Geuder r, T. Cebecauer f

aRenewable Energy Division, CIEMAT, Avda. Complutense, 40, 28040 Madrid, SpainbDLR, Institute of Solar Research, PSA, SpaincUniversity of Jaen, MATRAS Group, Spain

dSuntrace GmbH, GermanyeSolar Consulting Services, Spain

fGeoModel Solar, Slovakiag IrSOLaV, Spain

hMINES ParisTech, OIE, FranceiAustralian Bureau of Meteorology, Australia

jUniversity of South Australia, AustraliakUniversity of Geneva, Institute F.-A. Forel, Switzerland

lMeteotest, SwitzerlandmPontificia Universidad Catolica de Chile, Chile

nWorld Energy & Meteorology Council, UKoNREL, Boulder, USA

pDanish Meteorological Institute, DenmarkqDave Renne Renewables, LLC, USA

rCSP Services, University of Applied Sciences Stuttgart, Germany

Received 3 November 2015; received in revised form 29 February 2016; accepted 1 March 2016Available online 14 March 2016

Communicated by: Associate Editor Jan Kleissl

Abstract

At any site, the bankability of a projected solar power plant largely depends on the accuracy and general quality of the solar radiationdata generated during the solar resource assessment phase. The term ‘‘site adaptation” has recently started to be used in the frameworkof solar energy projects to refer to the improvement that can be achieved in satellite-derived solar irradiance and model data when short-term local ground measurements are used to correct systematic errors and bias in the original dataset. This contribution presents a pre-liminary survey of different possible techniques that can improve long-term satellite-derived and model-derived solar radiation datathrough the use of short-term on-site ground measurements. The possible approaches that are reported here may be applied in different

http://dx.doi.org/10.1016/j.solener.2016.03.001

0038-092X/� 2016 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +34 913466043; fax: +34 913466037.E-mail address: jesus.polo@ciemat.es (J. Polo).

26 J. Polo et al. / Solar Energy 132 (2016) 25–37

ways, depending on the origin and characteristics of the uncertainties in the modeled data. This work, which is the first step of a forth-coming in-depth assessment of methodologies for site adaptation, has been done within the framework of the International EnergyAgency Solar Heating and Cooling Programme Task 46 ‘‘Solar Resource Assessment and Forecasting”.� 2016 Elsevier Ltd. All rights reserved.

Keywords: Satellite-derived solar radiation; Site adaptation; Bankability of data for solar projects; Solar radiation model-derived data

1. Introduction

Accurate and precise knowledge of the local solarresource is a prerequisite for the successful deployment ofany solar energy system. For instance, reliable and accuratedata of different components of solar radiation are neededfor concentrating solar power (CSP) projects, which areprimarily interested in direct normal irradiance (DNI),and flat-plate collectors (PV or thermal) which requiregood predictions of the incident global horizontal (GHI)or global tilted irradiance (GTI), also referred to as ‘‘planeof array” (POA). From the initial site selection stage to theplant’s design and financing the solar resource assessmentplays a major role in the success of the project whenevermonthly or annual plant predictions are needed (Stoffelet al., 2010; Sengupta et al., 2015).

High-quality solar resource measurements at the targetsite are usually not available to ensure proper solarresource assessment and thus to secure the final acceptanceof the project. Additionally, the inter-annual variability ofsolar radiation components plays an important role in theuncertainty associated with the energy yield prediction ofsolar plants. As a consequence, long-term, typicallymulti-decadal, time series of the key solar radiation compo-nents are required at various stages of any solar plant pro-ject (Meyer et al., 2006; Stoffel et al., 2010). Since multi-decadal measurements are hardly available at any potentialsolar power plant site, long-term time series of solar irradi-ance must generally be supplied from modeled data, typi-cally using satellite imagery to derive the influence ofclouds on solar radiation at the Earth’s surface.

Solar radiation components at the Earth’s surface canbe determined from several methods and differentapproaches. The most widely known ones are those basedupon the use of meteorological satellites (Hammer et al.,2003; Mueller et al., 2003; Rigollier et al., 2004; Janjaiet al., 2005; Polo et al., 2008; Perez et al., 2013). The under-lying methodologies have been evolving during the last30 years, following advances in remote sensing and compu-tational techniques. The literature documents this trendthat started with simple methods based on a crude atmo-spheric energy balance (Gautier et al., 1980; Moser andRaschke, 1983; Cano et al., 1986) and has progressed tothe more intricate methods currently in use (Perez et al.,2002; Rigollier et al., 2004; Mueller et al., 2004; Schillingset al., 2004; Viana et al., 2011). For solar applications,most developments follow an empirical or semi-empirical

approach, also referred to as the cloud-index method.The common approach to determine DNI with the cloud-index method is the use of a GHI-to-DNI separationmethod (Gueymard and Ruiz-Arias, 2014). In recent years,a convergence between the physical and the semi-empiricalapproach has started to manifest, with semi-empiricalmodels tending to include more physical descriptions.Physical methods (using retrievals of cloud optical proper-ties from satellite observations) can now provide appropri-ate datasets for solar applications (Sengupta et al., 2015).

Satellite based methods are able to provide long-termhourly or even sub-hourly time series of solar radiationcomponents. A thorough validation exercise of most ofthe current satellite-based databases has been performedwithin the International Energy Agency (IEA) Solar Heat-ing and Cooling (SHC) Programme Task 46 ‘‘SolarResource Assessment and Forecasting” activities(Ineichen, 2014). Different comparative and evaluationstudies have shown that despite the high accuracy and reli-ability of satellite-derived data, significant differences fromground data may result (e.g., Gueymard, 2011; Ineichen,2014). This is a consequence of several sources of error,whose complex causes typically affect GHI, GTI andDNI differently (Cebecauer et al., 2011). For instance,satellite-derived methods usually require external atmo-spheric information on the most important attenuatingcomponents under cloudless conditions, namely aerosolsand precipitable water vapor, with an important impacton the final uncertainty (Gueymard and Thevenard, 2009;Gueymard, 2012, 2014). In addition, site-specific informa-tion (albedo or topography) is also needed and can affectthe final uncertainty. A complete list of known issues andpotential uncertainty sources in satellite-based solar modelsis outlined in previous works (Suri and Cebecauer, 2014).

For sound solar resource assessments, all model-derivedestimates must be validated and qualified as much as pos-sible using locally available ground measurements. More-over, all large solar power projects must respond tostringent bankability criteria, following the regular practiceof financial institutions. In this context, detailed and accu-rate knowledge of the seasonal and inter-annual variabilityin the solar resource is of crucial importance. Therefore,bankable energy production scenarios must be based on astatistical analysis of long-term time series. Since a largeinter-annual variability of the solar resource means a largerrisk, this variability must be thoroughly assessed. Currentstudies demonstrate that the inter-annual variability in

J. Polo et al. / Solar Energy 132 (2016) 25–37 27

DNI is significantly larger than that in GTI, and largerthan that in GHI (Lohmann et al., 2006; Gueymard andWilcox, 2011; Gueymard, 2012).

Therefore, the proper evaluation and characterization ofthe uncertainty sources embedded in satellite-derived dataconstitutes a crucial step toward the bankability of solarresource assessments for solar power plants. For instance,the proper identification and correction of systematicerrors or bias in satellite-derived data by comparison withhigh-quality ground data can lead to a large reduction ofuncertainty. More precisely, if the satellite-derived esti-mates for a specific site can be calibrated against a short-term local measurement campaign, the long-term solarresource accuracy can be substantially improved. Such anon-site campaign is normally required by financial institu-tions, and typically starts a few months after the beginningof the project.

This process of calibration or correction of modeleddata is similar to what has been developed for the windindustry in the past (Potter et al., 2008). Similar to the caseof wind energy resource assessment, different methods maybe used, which are discussed in the following sections.Whereas the purpose of these methods is generally agreed,the terminology varies depending on the method’s propo-nent. For instance, these methods may be referred to as‘‘site adaptation” (Suri and Cebecauer, 2011), ‘‘datasetmerging” (Thuman et al., 2012), or ‘‘measured recordextension” (Bender et al., 2011; Gueymard et al., 2012).For clarity and conciseness, all these methods will be col-lectively referred to as ‘‘site adaptation” in what follows.In summary, all the various statistical methods that havebeen developed to decrease the uncertainty in the localsolar resource try to improve satellite-derived irradiancedata (by lowering their random errors, and most impor-tantly their bias) using characteristics of correspondingground observations during overlapping time periods.

Within IEA-SHC’s Task 46, a considerable effort hasbegun to improve the bankability of solar radiation data-sets by developing procedures and methods designed toincrease the quality of both ground measurements andmodeled data, and to combine them as efficiently as possi-ble. In this sense, a thorough benchmarking exercise on siteadaptation techniques is being designed and will provideresults in the near future. This paper is a first step in explor-ing the issue of site adaptation for bankability and offers apreliminary, but needed, survey of the different techniquesfor site adaptation reported elsewhere. The informationhere gives a general overview on the current state andpoints out the further work to be done.

2. Importance of quality ground measurements

For an effective use of ground irradiance measurementsin improving and evaluating solar radiation datasets, thequality of the measurements must be assured to be as highas possible. All types of instruments should be properly cal-

ibrated, operated and well maintained, which in particularimplies regular and frequent cleaning. The measurementprotocol should be rigorous and follow the best practicesand available standards (Stoffel et al., 2010; Senguptaet al., 2015). In addition, the derived time series shouldbe thoroughly quality controlled, and the number of datagaps minimized. Quality check methods such as BSRN rec-ommendations (McArthur, 1998), SERI-QC (NREL,1993) or those delivered in the MESoR project (Hoyer-Klick et al., 2009) should be used to detect either valuesbeyond physical limits or inconsistent values as a conse-quence of errors in solar tracking systems. The use of suchmethods requires the availability of the three essential solarirradiance components: GHI, DNI and Diffuse horizontalirradiance (DHI).

Long-term datasets of solar irradiance components tobe used for bankability should be accompanied by theircorresponding uncertainty. Site adaptation methodsrequire high-quality ground solar radiation data, i.e. lowdata uncertainty, whose magnitude must also be assessed.

The highest accuracy in solar radiation measurement isnormally achieved by using thermopile radiometers forthe simultaneous measurement of GHI, DNI and DHI(Vignola et al., 2012). Among the thermopile-based instru-ments currently available, those denoted as ‘‘secondarystandards” have the highest precision (ISO, 1990). Second-ary standards, under typical operating conditions andproper maintenance, have a measurement uncertainty (at95% confidence) of 5% for GHI, 3% for DNI, and 7%for DHI (Vignola et al., 2012). The three main solar radia-tion components can be alternatively measured by a singlerotating shadowband radiometer (RSR), also referred to asrotating shadowband irradiometer (RSI). This device usesa photodiode sensor that is periodically shaded by a shad-owband that rotates across the horizontal detector. There-fore the RSI measures GHI when unshaded and DHI whenshaded, out of which DNI is appropriately derived by cal-culation from the other two components (Sengupta et al.,2015; Wilbert et al., 2015a). They usually require less main-tenance than thermopile-based instruments, which is animportant advantage in remote areas where soiling condi-tions could prevail. The proper calibration of RSIs involvesvarious steps and empirical corrections, the effectiveness ofwhich are still being investigated (Jessen et al., 2015). Theanalysis of long-term behavior of the accuracy of RSI typ-ically requires a recalibration of the device after two yearsof operation (Geuder et al., 2014). A thorough analysis ofaccuracy and uncertainties in broadband shortwaveradiometers is found in the literature (Vuilleumier et al.,2014). In addition, RSI uncertainty has been investigatedrecently with significant updates and improvements(Wilbert et al., 2015b).

As a summary of the existing literature, the stepsdescribed below need to be followed with care to ensurean optimum quality of ground measurements that areintended to be used in site adaptation techniques:

28 J. Polo et al. / Solar Energy 132 (2016) 25–37

1. Proper design of the ground station technical character-istics according to the intended aim of the data (datalogger, instrument selection for additional relevantparameters such as wind speed, temperature, groundingand shielding; also general aspects like stable instrumentmounts, security/fence, position of station without shad-ing, etc.).

2. Selection of radiometers having low uncertainty for thesolar component(s) under scrutiny, e.g., DNI and/orGTI taking also into account the available maintenance.

3. State-of-the-art calibration of the radiometers and anadequate schedule for future calibrations.

4. Regular maintenance of the station – mainly cleaning ofsensor heads and verification of alignment and level.

5. Commissioning and regular check of the station byqualified experts including proper documentation of allquality-related aspects.

6. Extensive quality control of the recorded data to assesstheir effective accuracy, flagging data, and applying cor-rections (only where advisable).

Only measurements fulfilling all these requirements canbe regarded as valid and can be used for fusion with, orcorrection of, model-derived datasets. Otherwise, the qual-ity of model-derived datasets could be lowered by mea-sured data of insufficient quality, due to, for example,significant bias. In addition, measurements of temperatureand wind speed observations at heights critical to theplanned solar system can be useful for the de-biasing ofancillary long term temperature and wind speed measure-ments that are also needed in many simulation tools.

3. Solar radiation estimation from satellite imagery

The use of meteorological satellites for deriving solarradiation components at specific sites or over vast regionsis nowadays the most usual approach for solar resourceassessment studies. These methodologies have been evolv-ing during the last 30 years, following advances in remotesensing and other techniques. The literature documents thistrend from the first simple methods (Gautier et al., 1980;Moser and Raschke, 1983; Cano et al., 1986) through tothe more intricate methods in current use (Perez et al.,2002; Rigollier et al., 2004; Mueller et al., 2004; Schillingset al., 2004; Janjai et al., 2005; Martins et al., 2007;Zarzalejo et al., 2009; Polo et al., 2012, 2014; Escobaret al., 2014, 2015).

Traditionally, there have been two main approaches forsatellite-derived data. Physical models use the satelliteinformation to solve the radiative transfer in the atmo-sphere and they usually retrieve vertically averaged proper-ties of clouds and cloud optical depth (Miller et al., 2013).On the other hand, empirical models are based on the sta-tistical relationship between spaceborne and ground obser-vations. In recent years, some convergence between thesetwo different approaches has started to manifest, so thatmost of the models in current use can be referred to as

semi-empirical models. In effect, they include analyticalmethods to describe the interaction of radiation with gasesand aerosols in the atmosphere, while using satellite-derived information to empirically retrieve the cloud index(Perez et al., 2013).

The high degree of maturity in models and methods forestimating solar radiation components from satellite ima-gery has resulted in the availability of many different data-bases and web services that provide time series of satellite-derived solar radiation data. Table 1 lists the most fre-quently used databases and summarizes some of their gen-eral characteristics. Note that some of these databases arein some way dependent on each other. Satellite-based data-bases and methods have been extensively assessed withground measurements elsewhere. Thorough reviews andcompilations of these validation results can be found inthe literature (Ineichen, 2011, 2014; Sengupta et al., 2015).

Time series derived from satellite observations offer thegreat advantage of providing global (worldwide) coverage(at least between latitudes of 60�S and 60�N), and longperiods of data (up to �20 years currently, for geographicareas covered by the oldest satellites with appropriate sen-sors). The current fleet of geostationary satellites is judi-ciously positioned to continuously monitor reflectedspectral radiances from the atmosphere over the wholeglobe (except high latitudes) at hourly or sub-hourly timesteps. The nominal spatial resolution of the satellite-derived irradiance data is in the range of 1–10 km. Thelarge geographical coverage of these methods means theyare able to supply solar irradiance information for almostevery site on Earth. However, shortcomings do existbecause satellite-based methods cannot always model allpossible local effects or specific features of a target site.Indeed, complex terrain with very variable elevation,coastal areas, areas or periods with unusual or rapidlychanging cloud conditions, highly reflective surfaces (typi-cal of sand or salt deserts and snow-covered areas), andinaccuracies in information on local atmospheric con-stituents (mostly aerosol optical depth, AOD, and watervapor), are examples of the possible features that mightsubstantially affect the final uncertainty, and hence thebankability of solar energy projects (Cebecauer and Suri,2010, 2012; Cebecauer et al., 2011). Thus in case of groundsnow cover under clear sky conditions the satellite modelsbased on the visible channel might result in erroneouscloudy estimations; the multichannel (visible and IR) satel-lite model techniques can offer better results (Perez et al.,2010). Similar situations can occur in bright desert regionswhere specific models for ground albedo could prevent thaterrors (Polo et al., 2012). This leads to temporary overesti-mations of cloud optical thicknesses or, cloud indices.Satellite model performance is also affected by the solarand viewing geometry. At high solar zenith angles highclouds cast shadows upon lower clouds that can be misin-terpreted as cloud free areas. At high view zenith angles aparallax effect occurs, where the apparent cloud positionis different from the actual position.

Table 1Databases and services providing solar radiation time series derived from satellite information.

Name Time basis Coverage Web site

NSRDB update 30-min USA http://rredc.nrel.gov/solar/old_data/nsrdb/NASA SRB 3-hourly World http://gewex-srb.larc.nasa.gov/DLR-ISIS 3-hourly World http://www.pa.op.dlr.de/ISIS/HelioClim hourly Europe–Africa http://www.soda-is.com/eng/helioclim/SOLEMI hourly Europe–Africa–Asia http://wdc.dlr.de/data_products/SERVICES/SOLARENERGY/SolarGIS 30-min World http://solargis.info/EnMetSol hourly Europe–Africa https://www.uni-oldenburg.de/en/physics/research/ehf/

energiemeteorology/enmetsol/IrSOLaV hourly Europe–Africa–Asia–

Americahttp://irsolav.com/

CM-SAF (SARAH) hourly Europe–Africa http://www.cmsaf.eu/SUNY – SolarAnywhere 30-min North America http://www.solaranywhere.com/MACC RAD 3-hourly World http://www.soda-pro.com/es/help/macc-rad/automatic-accessPVGIS hourly Europe–Africa–Asia http://re.jrc.ec.europa.eu/pvgis/Meteonorm TMY World http://www.meteonorm.com/CPTEC/INPE daily South–America http://satelite.cptec.inpe.br/radiacao/Vaisala hourly World http://www.vaisala.comAustralian Bureau of

Meteorologyhourly Australia http://www.bom.gov.au/climate/data-services/solar-information.shtml

J. Polo et al. / Solar Energy 132 (2016) 25–37 29

Advancement of satellite-derived methods for solar radi-ation computation typically results from improvements inthe design of satellite sensors and from improved atmo-spheric retrieval methods as well. Thus, new and better ver-sions of the satellite-derived databases and their underlyingmodels have appeared recently (Mueller et al., 2012; Amilloet al., 2014; Qu et al., 2014; Hammer et al., 2015; Perezet al., 2015). In addition, several methods are being opti-mized for working with different geostationary satelliteswithout the need to modify their fundamental methodol-ogy (Amillo et al., 2014; Albarelo et al., 2015).

4. Methodologies for site adaptation of satellite-derived data

The rapidly increasing deployment of solar power plantsover many different climatic zones in the world has encour-aged studies and attempts focused on improving the accu-racy of satellite estimations of solar radiation components.Most attention has been devoted to DNI because it is gen-erally determined with higher uncertainty than GHI. Thisresults from the much higher sensitivity of DNI to cloudi-ness and aerosols. Various possible site adaptation tech-niques exist and are reviewed below. They are all basedon the use of short-term ground measurements of solarirradiance and/or accurate atmospheric data at the siteunder scrutiny to ‘‘anchor” the large-scale modeledresource data.

4.1. Physically based methods

One approach for correcting model-derived solar radia-tion data consists of adjusting the atmospheric input dataso that the new results better match the ground-basedobservations. In principle, this approach might lead to amodification of any input, such as cloud properties, but

in most cases the main focus is on aerosol turbidity becauseof its large impact on DNI (Gueymard, 2012).

Many satellite derived hourly DNI time series areaffected by significant monthly and annual bias overregions where high aerosol loads are frequent (Cebecaueret al., 2011; Gueymard, 2011). Therefore, to obtain highaccuracy and low bias in solar irradiance estimates (partic-ularly DNI), AOD used as input in satellite-based methodsmust have the lowest bias possible (Ruiz-Arias et al.,2015a).

Clear sky transmittance models can be used withground-based AOD observations for the purpose of cor-recting pre-existing model estimates. As an example, amethod based on the REST2 model (Gueymard, 2008) isreported in the literature (Gueymard, 2012). However,due to the scarcity of ground AOD observations the correc-tion method was generalized to the use of large-scale AODdatasets. The methodology consists of comparing hourlyDNI and GHI outputs of the REST2 model with DNIand GHI derived from the satellite model for clear-sky con-ditions, Isat. A correction factor (RC) can be obtained andused to correct the satellite-based irradiance.

Rc ¼ IREST2

Isatð1Þ

Similar methods have been proposed that use otherclear-sky models to adapt the atmospheric aerosol datasetaccording to the irradiance measured during clear days(Cebecauer and Suri, 2012).

The generation of new and more accurate global aerosoldatasets is also an ongoing issue, to which important effortsare being devoted. For instance, Gueymard proposed aglobal aerosol dataset with 0.5� � 0.5� spatial resolutionusing monthly averages from MODIS retrievals, MATCHreanalysis and AERONET measurements, also introducingan altitude correction for AOD based on a Digital

30 J. Polo et al. / Solar Energy 132 (2016) 25–37

Elevation Model (DEM) to reduce systematic errors(Gueymard and Thevenard, 2009). This dataset was laterimproved by the addition of other sources of data, suchas MISR satellite retrievals and gridded climatology(Gueymard and Sengupta, 2013).

Similarly, Ruiz-Arias et al. (2013a, 2013b) proposed adata fusion approach of daily gridded AOD estimates fromMODIS retrievals and daily point-wise observations ofAOD from AERONET stations. The method is based ona combination of kriging and optimal interpolation meth-ods and was tested in the continental United States. First,the missing satellite AOD estimates over cloudy areas werefilled using kriging techniques. Second, the bias at regionalscale was removed using a correlation model of this biasand the observed AOD MODIS values (Ruiz-Arias et al.,2013a). Third, the debiased AOD estimates were locallyadjusted to match the AERONET aerosol observationsusing optimal interpolation. Therefore, the method pro-poses an approach that corrects the satellite AOD errorssuccessively at regional and local scales. In contrast withother methods introduced above, the local adjustment doesnot correct the satellite estimates only where a collocatedAERONET sunphotometer exists. In addition, it uses thecovariance of the AOD error between neighboring sitesin such a way that the AOD satellite error at the AERO-NET locations is cast over surrounding areas. In thisway, the area of influence of the ground AOD observationsis widened.

A DNI sensitivity analysis of this method conductedunder clear-sky conditions showed that, using the originalAOD satellite retrievals, filled by interpolation, the result-ing DNIs for AOD below 0.2 are overall positively biasedby more than 5%, with a significant spread over theobserved value. If, in contrast, corrected AOD values areused, the overall DNI bias vanishes and 90% of the DNIestimates have an error lower than 5% (Ruiz-Arias et al.,2013b). These results were applied to North America,where AOD is generally low to moderate. Much largeraerosol loads are frequent over many regions of the world,where similar studies would thus be necessary.

Calibration of MODIS L3 daily values of both AODand precipitable water have been recently proposed byusing linear regression with AERONET measurementsshowing an improved performance of REST2 clear skymodel (Zhong and Kleissl, 2015).

4.2. Statistical methods

Statistical adaptation of model-derived results is fre-quently used in various fields of applied meteorology.For instance, rain rates can be corrected to better reflectlocal conditions. Moreover, it is also quite common toadapt wind speeds to local measurements for wind resourceanalysis. Similarly, statistical methods can be applied toadjust model/satellite-derived solar radiation values tolocal measurements. Several such approaches are currentlyused in practice, and are described in the next subsections.

4.2.1. Bias removal by linear adaptation

Satellite-based methods for deriving solar radiationcomponents may exhibit systematic errors that could yieldan overall overestimation or underestimation trend,referred to as bias, and characterized by the mean biasdeviation (MBD) statistic. Bias between the predicted andmeasured irradiance may result from systematic featuresof the radiative model or from regional inconsistencies inthe external input data (particularly aerosols or watervapor).

When short-term high quality ground DNI and GHImeasurements are available bias in satellite irradiance esti-mates can be reduced by calculating correction factorsfrom the overlapping data (Carow, 2008; Cebecauer andSuri, 2010; Vindel et al., 2013; Harmsen et al., 2014).

The simplest method is to estimate the MBD andremove it from the whole dataset. A bias removal can alsobe developed by fitting a line to the cloud of points in ascatter plot and subtracting the y ¼ x line. Fig. 1 illustratesan example of this simple method (Polo et al., 2015). Cor-rection by the difference between the fitted line(ysat ¼ axground þ b) and the y ¼ x line produces new esti-mates with negligible bias. The new estimates are computedby the following resulting equation:

ynew ¼ ysat � ½ða� 1Þxground þ b� ð2ÞIn many situations the linear fit corrections should be

applied to a subset of derived data rather than to the wholedataset, for instance clear sky bias corrections (Kankiewiczet al., 2014). Likewise a seasonal method was proposed toadapt satellite estimates of DNI over India, where initialdeviations were typically large and seasonal (Polo et al.,2015).

4.2.2. Non-linear methods

Some proposed methods introduce a parameter-dependent correction, others a so-called ‘‘feature transfor-mation” correction (Carow, 2008). The first method uses aspecific combination of parameters (for example clear-skyindex or sun elevation angle), and additively or multiplica-tively modifies the original satellite-based time series. Typ-ically, the additive correction efficiently reduces MBD andthe Root Mean Squared deviation (RMSD) on average.The second approach, feature transformation, uses proper-ties of the cumulative distribution function of the on-siteground measurements and transfers them via look-uptables to the satellite-derived time series. This processbrings the distribution functions closer together. It hasbeen shown that improvements made by feature transfor-mation strongly depend on the length of the overlap periodwhen both satellite-derived irradiance data and ground-based measurements are available (Schumann et al.,2011). It is recommended to take at least a whole year ofmeasurements to develop this adaptation algorithm.

The literature reports another method based on theassumption that low irradiancevaluesareoftenoverestimated,whereashigh irradiancevalues tend tobeunderestimated, thus

Fig. 1. Example of linear bias removal for DNI data for 2001 in Tabernas (Spain) according to Eq. (2). Uncorrected data on the left and corrected data onthe right. Ground data from DLR-PSA station (Kipp and Zonen CH1 (until 30/8/2011) and CHP1 (afterward)). Satellite data estimated by CIEMATprocedure from Meteosat Second Generation imagery.

J. Polo et al. / Solar Energy 132 (2016) 25–37 31

leading to an artificial compensation in terms of the yearlyirradiation sum. A third-degree polynomial is then assumedto correct the data (Mieslinger et al., 2013).

Other reported methods of combining short-termground measurements with longer-term satellite data arebased on model output statistics (MOS) (Bender et al.,2011). MOS is a multi-variate linear regression analysisbetween a prescribed set of predictors (satellite deriveddata or estimates from numerical weather prediction mod-els, for instance) and the surface observational dataset. Theobjective is to develop a multi-linear regression equation tocorrect the estimated values, thus effectively removing thebias and adjusting the variance. An example of MOS cor-rection applied to DNI data from Tamanrasset, Algeria,evaluated the impact of varying overlapping periods ofmeasured and modeled data, from 0 to 12 months in 3-month increments and showed how the bias was being low-ered as a function of the overlap length (Gueymard et al.,2012). As expected, the longer the overlapping period thelower the uncertainty. However, an interesting result isthat, for some of the five sites tested (which encompassedvery low and stable to very high and variable AOD condi-tions), the improvement in modeled irradiance biasbecomes modest for overlapping DNI observation periodslonger than about nine months, particularly if an uncer-tainty of 5% is considered sufficient. This seems to indicatethat the method is very efficient and converges rapidly. Ashort overlapping period is obviously desirable since itaccelerates the final steps of the solar project preparation.

As with all other adaptation methods, the availability ofhigh-quality ground data is crucial to the success of MOS,since the observations have a direct impact on how well thestatistical model behaves for local conditions. A methodsimilar to MOS, known as Measure–Correlate–Predict(MCP), is also reported (Thuman et al., 2012).

A novel method based on Fourier decomposition hasbeen proposed to calibrate the daily global irradiationestimated by the HelioClim-3 database (Vernay et al.,

2013). The method consist of performing a Fouriertransform on the clearness index errors and developing aregression to correct the daily clearness index, KT, such as

KT�HC3 ¼ aST þ dSTKTHC3 þ bST cosð2pFdjÞ þ cST sinð2pFdjÞ

ð3Þwhere KT is the daily clear index, F is the frequency(1/365 day�1) and dj is the day number of the year. Thismethodology effectively calibrates the satellite-deriveddaily global irradiation through the use of one year ofon-site measurement campaign. Moreover, the authorshave extended the method to allow the use of shorter on-site campaigns. For instance, for the Provence–Alpes–Coted’Azur (PACA) region in southeast France, they concludethat a 9-month campaign starting between January andMay can give good accuracy. The authors have also shownthat good results could be obtained when extending thiscalibration method to other European and African sitesusing mainly BSRN stations (Vernay et al., 2012).

4.2.3. Regional site adaptation

The methods presented so far make use of an overlap-ping period of concurrent ground and model-derived datato develop a correction for the latter over its entire record,thus including non-overlapping periods. Hence, the appli-cation of this methodology is possible only if overlappingmodeled data and ground-based observations exist, for atleast 9–12 months, in order to include seasonal effects. Inaddition, the likely decrease in temporal auto-correlationas the time lag between model’s training and correctionperiods increases is not generally considered, which canresult in over- or under-corrections. This fact is particularlycritical for DNI, in which inter-annual variability is usuallymuch higher than for GHI (Lohmann et al., 2006; Pozo-Vazquez et al., 2011; Gueymard et al., 2012). In this sense,and in order to extend the availability of measurements,some authors have proposed methods that make use of

32 J. Polo et al. / Solar Energy 132 (2016) 25–37

irradiance observations recorded at one or more locationsclose to the target site. This approach is valid inasmuchas the spatial covariance structure between the model-derived data and ground measurements is considered(Ruiz-Arias et al., 2013b).

Recently, regional fusion methods for model-derivedsolar radiation data and ground measurements have beenpresented (Wald et al., 2003; Skamarock et al., 2008;Journee et al., 2012). Also recently, Ruiz-Arias et al.(2015b) have proposed an advanced regional fusionmethod, consisting of a numerical process by which themodeled solar radiation data (obtained typically fromsatellite-based techniques or numerical weather prediction(NWP) models, and provided as two-dimensional geo-referenced grids) are objectively adjusted at each modelgrid cell as a function of the reliability of the modeled solarradiation value at that grid cell with respect to the reliabil-ity of the nearby ground observations cast onto that gridcell. The feasibility of the method is demonstrated usingmonthly-averaged values of daily GHI and daily DNIobtained with the Weather Research and Forecasting(WRF) NWP model (Skamarock et al., 2008). Althoughthe correction method is demonstrated only for NWP-based solar radiation estimates, its authors state that itwould be equally applicable to satellite-derived solar radi-ation gridded datasets and other time scales.

4.2.4. Site adaptation using the cumulative distribution

function

Another type of correction method focuses on fittingand improving the cumulative distribution function(CDF) of satellite-derived data by comparison with the

Fig. 2. Site adaptation based on fitting

CDF of ground data (Cebecauer and Suri, 2012). TheEuropean ENDORSE project also investigated the appli-cation of a feature transformation using the difference ofdata sets in cumulative distribution functions (Blancet al., 2012). The feature transformation is based on theadaptation of the frequency distribution of the modeleddata to that of the ground data by using information ofthe relative distortion of the distribution of the satellitedata from parallel sets (Carow, 2008; Schumann et al.,2011). The correction for a given satellite irradiance valueis then defined as the difference between the ground-measured irradiance with the same CDF as the satelliteirradiance and the satellite-derived irradiance itself. Themain objective of this kind of method is to find a way totransform the estimated data based on the approximationof the CDF of satellite-derived dataset into the grounddata CDF, as illustrated in Fig. 2.

The impact of the overlap period on the improvement ofsatellite-derived data has also been studied with the featuretransformation method (Schumann et al., 2011). In somecases these authors found that three continuous monthsof ground data could largely improve the modeled dataset.However, they stated that in order to take into account dif-ferent seasonal characteristics at least one year of grounddata would be needed.

4.2.5. Adaptation of input parameters of a satellite-based

solar radiation model

An alternative approach for site adaptation consists ofcorrecting the input parameters of the satellite modelinstead of working on the solar radiation derived values.Two methods can be found in the literature (Cebecauer

cumulative distribution functions.

J. Polo et al. / Solar Energy 132 (2016) 25–37 33

and Suri, 2015): (i) adaptation of clearness index; and (ii)adaptation of atmospheric input data. The clearness indexintegrates into a single number all properties of the atmo-sphere (effects of aerosols, water vapor and clouds) thateventually attenuate solar irradiance. This correction maybe applied to high-frequency data (hourly or sub-hourly)or to daily data. It was found, however, that more accurateand consistent GHI and DNI results were obtained withmethods based on the adaptation of the model’s input data,combined with the control of parameters affecting the satel-lite model, and a subsequent recalculation of the satellitemodel over the whole long-term period. Hourly (or sub-hourly) data are typically used. Both approaches generallyfollow three main steps:

� Use of a solar radiation model and ground measure-ments to derive the primary model input parameters.

� Calibration/correlation of the input parameters by aleast-square regression (or similar method) of hourly(or daily) difference between the input parametersderived from the ground measurements and the corre-sponding satellite-based parameters.

� Recalculation of GHI, DIF and DNI, with the solarradiation model now using site-adapted inputs.

Fig. 3 illustrates an example of the correction of AODderived from MACC retrievals (Inness et al., 2012) usingground data from an AERONET sunphotometer station.

4.3. MCP applied to satellite and re-analysis data

Measure–Correlate–Predict (MCP) methods have beenextensively used to estimate the wind resources that

Fig. 3. Example of site adaptation on

represent the long-term conditions at a target site where ashort-term wind data measurement campaign has beenheld (Carta et al., 2013). The results of an exercise per-formed within the activities of the IEA SHC Task 46 forexploring the use of measure–correlate–predict (MCP)techniques applied to satellite and re-analysis data is pre-sented in this section. The following MCP method isapplied to yearly averages, assuming that the relationshipbetween a shorter time period and a longer time periodstays constant between the different datasets or nearby sta-tions or grid points:

Gy;l ¼ Gx;l

Gx;s� Gy;s ð4Þ

where Gy,l is the long term average at location y, Gx,s is theshort term (e.g., one year) average at location x, Gx,l is thelong term average at location x and Gy,s is the short term(one year) average at location y.

The methodology has been tested with the followingdatasets:

� NASA Modern-Era Retrospective Analysis forResearch and Applications (MERRA), re-analysis, reso-lution: 0.5 � 0.67� (http://gmao.gsfc.nasa.gov/research/merra/).

� NOAA NCEP, re-analysis, resolution: 1.5 � 1.5�(http://www.ncep.noaa.gov/).

� ECMWF ERA-INTERIM, re-analysis, resolution:0.75 � 0.75� (http://apps.ecmwf.int/datasets/data/in-terim_full_daily/).

� Helioclim3.0 (HC) dataset (Version 5), satellite data,resolution �5 � 5 km (http://www.soda-i-s.com/eng/he-lioclim/helioclim3_versions.html).

model input: correction of AOD.

34 J. Polo et al. / Solar Energy 132 (2016) 25–37

An assessment of the results has been performed withirradiance observations from stations of the Baseline Sur-face Radiation Network (BSRN – http://bsrn.awi.de/);Tamanrasset and Carpentras), MeteoSwiss (Bern/Zol-likofen, Locarno/Magadino and Zurich/Fluntern) and 22sites of Global Energy Balance Archive (GEBA – http://www.geba.ethz.ch/). For BSRN and MeteoSwiss the 10-year averages for 2005–2014 are used, whereas for GEBAthe 30-year period 1981–2010 is generally used.

Additionally, long-term trends of MERRA and ERAmodeled estimates, and of GEBA ground measurementshave been compared at 22 sites including the followingtests:

� comparison of linear long-term trends,� significance of trends (95% levels),� relative standard deviation of yearly averages (12monthly moving averages).

As shown in previous studies the bias of re-analysis datais high and is especially much higher than for satellite data(Boilley and Wald, 2015). For this reason re-analysis datashould not be used without corrections. The results of theoriginal RMSD values and the results based on MCP arelisted in Table 2. The simple MCP method reduces theRMSD of re-analysis data by 60–70%, compared to 20%for the satellite data. MCP applied to satellite data leadsto the best results (RMSD: 3.0%). MCP with re-analysisdata results in only slightly higher uncertainties (3.3% forMERRA and 3.4% for ERA). Only the uncertainty inthe MCP-corrected NCEP re-analysis is clearly higher(6.8%). Note that these RMSD values represent the year-to-year variability in the specific case of a 1-year measure-ment campaign. This shows that even though irradiancedata derived from re-analysis tend to have higher uncer-tainties than those derived from satellite observations, theycan still be useful for long-term approximations afterproper MCP correction.

4.4. Combining satellite and weather model data

Improving the accuracy of satellite-derived solar irradi-ance can also be achieved by including the output of aNWP model. A study showed that using a nonparametricregression, satellite-derived global horizontal irradianceand direct normal irradiance can be improved by combin-ing them with irradiance that has been dynamically down-scaled using a NWP model. Irradiance, solar zenith angle

Table 2Relative RMSD of the estimation of the long term averages for the 2BSRN and 3 MeteoSwiss sites.

Dataset Original RMSD (%) MCP RMSD (%)

Re-analysis NCEP 20.3 6.8Re-analysis MERRA 16.3 3.3Re-analysis ERA 11.2 3.4Satellite (Helioclim) 3.8 3.0

and their interaction terms are taken as inputs to general-ized additive models (GAM) using smoothing splines.GAMs are preferred over widely used fourth-order polyno-mial models after testing their relative performance. Theaddition of NWP-derived irradiance as a GAM predictoris shown to reduce the RMSE (with ground measurementsserving as the point of truth) by a few percent, dependingon location (Troccoli, 2015).

5. Discussion and conclusions

Different characteristics of a solar power project deter-mine whether or not it can be funded by financial institu-tions at an acceptable and quantifiably small risk level.This bankability of a solar power project is strongly depen-dent on the solar resource assessment that is typically con-ducted at the onset of the project. Solar resourceassessment studies have become a standard practice ofthe industry. A proper solar resource assessment is requiredfor the identification of the best possible site. The assess-ment is normally dependent on a regionally validatedsatellite-derived solar radiation dataset, which in turn mustbe locally validated by on-site ground irradiance measure-ments of high quality.

Satellite-derived solar radiation datasets are necessaryfor site selection, solar resource assessment, power produc-tion projections, and system performance studies. How-ever, satellite-derived datasets have uncertainties thatoriginate from different sources and from the approxima-tions adopted by the models. These uncertainties can besignificantly reduced with the help of high-quality on-siteirradiance observations spanning a minimum period of 9–12 months. The present literature review shows that vari-ous methods and approaches have been explored so farto properly combine ground radiation data with modeledradiation data, with the general goal to improve the accu-racy of long-term satellite-derived datasets. In the solarenergy world this process is often referred to as ‘‘siteadaptation”.

The adjustment aims primarily at removing the annualand/or seasonal bias in the long-term modeled irradianceestimates. To that end, the necessary correction methodol-ogy can follow either one of two possible avenues. The firstavenue consists in adapting the input data ingested by themodel to better fit the local irradiance measurements. Inthe second avenue, empirical adjustments of the model out-put estimates are made by comparison with the measure-ments. Some of the empirical correction methods onlymodify the bias, whereas others also adjust the frequencydistribution of the irradiance values. In some situations,correcting the bias and reducing the dispersion ofsatellite-derived data result in improvement of the cumula-tive distribution function. In most situations, however, it islikely that no unique empirical method can cover all thepossibilities and thus can be successfully applied to thewhole dataset. Consequently, each site would likely requirea specific initial study to identify the sources of discrepancy

J. Polo et al. / Solar Energy 132 (2016) 25–37 35

and then help design a proper method for data adaptation.The site specific method is often a combination of the dif-ferent approaches presented in this paper.

The optimum duration of the overlapping periodbetween ground observations and model estimates hasnot been widely studied so far. In particular, the effectsof multi-annual trends and inter-annual variability in theon-site measurements can potentially have large effects onthe radiation values and cumulative distribution functions.So far, such dynamic effects have not been fully covered inany of the different site adaptation methodologies that havebeen reviewed here. Until additional knowledge is availableon this topic, it is recommended that at least one year ofhigh-quality on-site irradiance measurements be used forsite adaptation, in order to cover the seasonal variabilityof solar radiation.

All site adaptation procedures reviewed here requirehigh-quality on-site data, which in turn indicates that anyshort-term measurement campaign must follow existingstandards and recommended best practice for the design,installation, operation and maintenance of all sensors, aswell as for the a posteriori quality control and data pro-cessing phases. Since a successful site adaptation procedurecritically depends on the quality of the on-site data, the uti-lization of low-quality data from substandard local groundmeasurement operations can lead to increased uncertaintyof the modeled datasets, erroneous plant sizing and yieldassessments. Such adverse results may in turn translate intoquestionable bankability and consequently increase thefinancial cost of the project, with the risk of making itunfeasible.

The uncertainty in long-term average solar irradiance isa dominating parameter in risk analysis of solar power pro-jects. The calculation of this uncertainty needs to take intoaccount the different uncertainties associated with themodel used to generate the long-term dataset and thoseassociated the on-site measurements used to validate andcorrect the modeled data. Therefore, a clear and systematicdefinition of the uncertainty associated with measurement-adjusted modeled data is strongly needed for bankablesolar resource assessments.

Further studies are needed to evaluate how well the var-ious algorithms presented here perform, depending on theoverlap duration and on the atmospheric, topographicand environmental conditions of various regions. Sincethe performance of the site adaptation process alsodepends on the quality of the satellite- or model-deriveddataset, other studies would have to investigate the impactof various satellite-derived datasets at many sites in widelydifferent regions. In this sense, within the IEA communityof solar resource experts it is planning to perform a deepbenchmarking exercise of site adaptation methods that isintended to provide guidance on the choice of methoddepending on the local conditions. Finally, it is remarkedthat the standardization of both site adaptation proceduresand data quality control methods for on-site measurementswould also contribute to a desirable harmonization of solar

resource datasets, leading to a smoother development ofsolar power projects and a stronger solar industry.

Acknowledgements

The authors wish to acknowledge the Solar Heating andCooling (SHC) Programme within the InternationalEnergy Agency (IEA), and in particular all of the expertswho are participating in Task 46 ‘‘Solar Resource Assess-ment and Forecasting”. This paper has been preparedwithin the framework of IEA SHC Task 46.

References

Albarelo, T., Marie-Joseph, I., Primerose, A., Seyler, F., Wald, L.,Linguet, L., 2015. Optimizing the Heliosat-II Method for surface solarirradiation estimation with GOES images. Can. J. Rem. Sens. 41, 86–100. http://dx.doi.org/10.1080/07038992.2015.1040876.

Amillo, A., Huld, T., Mueller, R., 2014. A new database of global anddirect solar radiation using the eastern meteosat satellite, models andvalidation. Rem. Sens. 6, 8165–8189. http://dx.doi.org/10.3390/rs6098165.

Bender, G., Davidson, F., Eichelberger, F., Gueymard, C.A., 2011. Theroad to bankability: improving assessments for more accurate financialplanning. In: Proceedings of: Solar 2011 Conf. American Solar EnergySoc., Raleigh, NC.

Blanc, P., Boilley, Alexandre, Killius, N., Massip, P., Schroedter-Hom-scheidt, Marion, Wald, L., 2012. Methods for data fusion. Report ofthe Project ENDORSE, 7th Framework Programme EU.

Boilley, A., Wald, L., 2015. Comparison between meteorological re-analyses from ERA-Interim and MERRA and measurements of dailysolar irradiation at surface. Renewable Energy 75, 135–143. http://dx.doi.org/10.1016/j.renene.2014.09.042.

Cano, D., Monget, J.M., Aubuisson, M., Guillard, H., Regas, N., Wald,L., 1986. A method for the determination of the global solar radiationfrom meteorological satellite data. Sol. Energy 37, 31–39.

Carow, J., 2008. Anpassung Langjahriger Satelliten-Strahlungszeitreihenan Bodenmesswerte Eines Jahres Diploma Thesis. University ofApplied Science, Nordhausen.

Carta, J.A., Velazquez, S., Cabrera, P., 2013. A review of measure–correlate–predict (MCP) methods used to estimate long-term windcharacteristics at a target site. Renew. Sustain. Energy Rev. 27, 362–400. http://dx.doi.org/10.1016/j.rser.2013.07.004.

Cebecauer, T., Suri, M., 2010. Accuracy improvements of satellite-derivedsolar resource based on GEMS re-analysis aerosols. In: Proceedings of:SolarPACES 2010 Conf., Perpignan, France.

Cebecauer, T., Suri, M., 2012. Correction of satellited-derived DNI timeseries using locally-resolved aerosol data. In: Proceedings of:SolarPACES 2012 Conf., Marrakech, Morocco.

Cebecauer, T., Suri, M., 2015. Site-adaptation of satellite-based DNI andGHI time series: overview and SolarGIS approach. In: Proceedings of:SolarPACES 2015 Conf., Cap Town, South Africa.

Cebecauer, T., Suri, M., Gueymard, C.A., 2011. Uncertainty sources insatellite-derived direct normal irradiance: how can prediction accuracybe improved globally. In: Proceedings of: SolarPACES 2011 Conf.,Granada, Spain.

Escobar, R.A., Cortes, C., Pino, A., Pereira, E.B., Martins, F.R.,Cardemil, J.M., 2014. Solar energy resource assessment in Chile:satellite estimation and ground station measurements. RenewableEnergy 71, 324–332. http://dx.doi.org/10.1016/j.renene.2014.05.013.

Escobar, R.A., Cortes, C., Pino, A., Salgado, M., Pereira, E.B., Martins,F.R., Boland, J., Cardemil, J.M., 2015. Estimating the potential forsolar energy utilization in Chile by satellite-derived data and groundstation measurements. Sol. Energy 121, 139–151. http://dx.doi.org/10.1016/j.solener.2015.08.034.

36 J. Polo et al. / Solar Energy 132 (2016) 25–37

Gautier, C., Diak, G., Masse, S., 1980. A simple physical model toestimate incident solar radiation at the surface from GOES satellitedata. J. Appl. Meteorol. 19, 1005–1012.

Geuder, N., Affolter, R., Kraas, B., Wilbert, S., 2014. Long-termbehavior, accuracy and drift of LI-200 pyranometers as radiationsensors in Rotating Shadowband Irradiometers (RSI). Energy Proce-dia 49, 2330–2339. http://dx.doi.org/10.1016/j.egypro.2014.03.247.

Gueymard, C., Sengupta, M., 2013. Improved Gridded Aerosol Data forIndia. NREL/TP-5D00-58762, National Renewable Energy Labora-tory, Golden CO (Report).

Gueymard, C.A., 2011. Uncertainties in modeled direct irradiance aroundSahara as affected by aerosols: are current datasets of bankablequality? ASME J. Solar Energy Eng. 133, 031024.

Gueymard, C.A., 2012. Temporal variability in direct and globalirradiance at various time scales as affected by aerosols. Sol. Energy86, 3544–3553.

Gueymard, C.A., Gustafson, W.T., Bender, G., Etringer, A., Storck, P.,2012. Evaluation of procedures to improve solar resource assessments:optimum use of short-term data from a local weather station to correctbias in long-term satellite derived solar radiation time series. In:Proceedings of: World Renewable Energy Forum, Denver CO.

Gueymard, C.A., Ruiz-Arias, J.A., 2014. Performance of separationmodels to predict direct irradiance at high frequency: validation overarid areas. In: Proceedings of: Eurosun Conference, Aix-les-Bains,France, 16–19 September.

Gueymard, C.A., 2008. REST2: high-performance solar radiation modelfor cloudless-sky irradiance, illuminance, and photosynthetically activeradiation – validation with a benchmark dataset. Sol. Energy 82, 272–285. http://dx.doi.org/10.1016/j.solener.2007.04.008.

Gueymard, C.A., 2014. Impact of on-site atmospheric water vaporestimation methods on the accuracy of local solar irradiance predic-tions. Sol. Energy 101, 74–82. http://dx.doi.org/10.1016/j.solener.2013.12.027.

Gueymard, C.A., Thevenard, D., 2009. Monthly average clear-skybroadband irradiance database for worldwide solar heat gain andbuilding cooling load calculations. Sol. Energy 83, 1998–2018.

Gueymard, C.A., Wilcox, S.M., 2011. Assessment of spatial and temporalvariability in the US solar resource from radiometric measurementsand predictions from models using ground-based or satellite data. Sol.Energy 85, 1068–1084. http://dx.doi.org/10.1016/j.solener.2011.02.030.

Hammer, A., Heinemann, D., Hoyer, C., Kuhlemann, R., Lorenz, E.,Muller, R., Beyer, H.G., 2003. Solar energy assessment using remotesensing technologies. Remote Sens. Environ. 86, 423–432. http://dx.doi.org/10.1016/S0034-4257(03)00083-X.

Hammer, A., Kuhnert, J., Weinreich, K., Lorenz, E., 2015. Short-termforecasting of surface solar irradiance based on Meteosat-SEVIRI datausing a nighttime cloud index. Rem. Sens. 7, 9070–9090. http://dx.doi.org/10.3390/rs70709070.

Harmsen, E.W., Tosado-Cruz, P., Mecikalski, J.R., 2014. Calibration ofselected pyranometers and satellite derived solar radiation in PuertoRico. Int. J. Renewable Energy Technol. 5, 43–54.

Hoyer-Klick, C., Beyer, H.G., Dumortier, D., Schroedter Homscheidt,M., Wald, L., Martinoli, M., Schillings, C., Gschwind, B.t., Menard,L., Gaboardi, E., Polo, J., Cebecauer, T., Huld, T., Suri, M., de Blas,M., Lorenz, E., Kurz, C., Remund, J., Ineichen, P., Tsvetkov, A.,Hofierka, J., 2009. MESoR – Management and exploitation of solarresource knowledge. In: Proceedings of SolarPACES 2009, Berlin.

Ineichen, P., 2011. Five satellite products deriving beam and globalirradiance validation on data from 23 ground stations. Report fromUniversity of Geneva, International Energy Agency.

Ineichen, P., 2014. Long term satellite global, beam and diffuse irradiancevalidation. Energy Procedia 48, 1586–1596. http://dx.doi.org/10.1016/j.egypro.2014.02.179.

Inness, A., Baier, F., Benedetti, A., Bouarar, I., Chabrillat, S., Clark, H.,Clerbaux, C., Coheur, P., Engelen, R.J., Errera, Q., Flemming, J.,George, M., Granier, C., Hadji-Lazaro, J., Huijnen, V., Hurtmans, D.,Jones, L., Kaiser, J.W., Kapsomenakis, J., Lefever, K., Leitao, J.,

Razinger, M., Richter, A., Schultz, M.G., Simmons, A.J., Suttie, M.,Stein, O., Thepaut, J.N., Thouret, V., Vrekoussis, M., Zerefos, C.theMACC team, 2012. The MACC reanalysis: an 8-yr data set ofatmospheric composition. ACPD 12, 31247–31347.

ISO, 1990. Specification and classification of instruments for measuringhemispherical solar and direct solar radiation. Edited by: InternationalOrganization for Standardization. ISO 9060.

Janjai, S., Laksanaboonsong, J., Nunez, M., Thongsathitya, A., 2005.Development of a method for generating operational solar radiationmaps from satellite data for a tropical environment. Sol. Energy 78,739–751. http://dx.doi.org/10.1016/j.solener.2004.09.009.

Jessen, W., Wilbert, S., Nouri, B., Geuder, N., Fritz, H., 2015. Calibrationmethods for rotating shadowband irradiometers and evaluation ofcalibration duration. Atmos. Meas. Tech. Discuss. 8, 10249–10282.http://dx.doi.org/10.5194/amtd-8-10249-2015.

Journee, M., Muller, R., Bertrand, C., 2012. Solar resource assessment inthe Benelux by merging Meteosat-derived climate data and groundmeasurements. Sol. Energy 86, 3561–3574. http://dx.doi.org/10.1016/j.solener.2012.06.023.

Kankiewicz, A., Dise, J., Wu, E., Perez, R., 2014. Reducing solar projectuncertainty with an optimized resource assessment tuning methodol-ogy. In: Proceedings of American Solar Energy Society (ASES)conference, San Francisco, USA.

Lohmann, S., Schillings, C., Mayer, B., Meyer, R., 2006. Long-termvariability of solar direct and global radiation derived from ISCCPdata and comparison with reanalysis data. Sol. Energy 80, 1390–1401.http://dx.doi.org/10.1016/j.solener.2006.03.004.

Martins, F.R., Pereira, E.B., Abreu, S.L., 2007. Satellite-derived solarresource maps for Brazil under SWERA project. Sol. Energy 81, 517–528. http://dx.doi.org/10.1016/j.solener.2006.07.009.

McArthur, L.J.B., 1998. Baseline Surface Radiation Network (BSRN).Operations manual V1.0. Serie: World Climate Research Programme.Secretariat of the World Meteorological Organization, Geneva,Switzerland.

Meyer, R., Lohmann, S., Schillings, C., Hoyer, C., 2006. In: Dunlop, E.D., Wald, L., Suri, M. (Eds.), Climate Statistics for Planning andSiting of Solar Energy Systems: Long-Term Variability of SolarRadiation Derived from Satellite Data. Nova Science Publishers/Earthlink.

Mieslinger, T., Ament, F., Chhatbar, K., Meyer, R., 2013. A new methodfor fusion of measured and model-derived solar radiation time-series.In: Proceedings of: SHC 2013, International Conference on SolarHeating and Cooling for Buildings and Industry, Freiburg.

Miller, S.D., Heidinger, A.K., Sengupta, M., 2013. Physically basedsatellite methods. In: Klessi, Jan (Ed.), Solar Energy Forecasting andResource Assessment. Elsevier (chapter 3).

Moser, W., Raschke, E., 1983. Mapping of global radiation and ofcloudiness from METEOSAT image data. Theory and ground truthcomparisons. Meteorol. Rdsch. 36, 33–41.

Mueller, R., Dagestad, K., Kuhlemann, R., Olseth, J., Reise, C.,Schroedter, M., Wald, L., Heinemann, D., 2003. Rethinking satellitebased solar irradiance modelling. In: Proceedings of: The 2003EUMETSAT Meteorological Satellite Data Users’ Conference,Weimar, Germany.

Mueller, R.W., Dagestad, K.F., Ineichen, P., Schroedter-Homscheidt, M.,Cros, S., Dumortier, D., Kuhlemann, R., Olseth, J.A., Piernavieja, G.,Reise, C., 2004. Rethinking satellite-based solar irradiance modelling:the SOLIS clear-sky module. Remote Sens. Environ. 91, 160–174.http://dx.doi.org/10.1016/j.rse.2004.02.009.

Mueller, R., Behrendt, T., Hammer, A., Kemper, A., 2012. A newalgorithm for the satellite-based retrieval of solar surface irradiance inspectral bands. Rem. Sens. 4, 622–647.

NREL, 1993. User’s Manual for SERI_QC Software. Assessing theQuality of Solar Radiation Data. National Renewable Energy Lab-oratory, Golden, CO (Report).

Perez, R., Cebecauer, T., Suri, M., 2013. Semi-empirical satellite models.In: Kleiss, Jan (Ed.), Solar Energy Forecasting and Resource Assess-ment. Elsevier (chapter 2).

J. Polo et al. / Solar Energy 132 (2016) 25–37 37

Perez, R., Ineichen, P., Moore, K., Kmiecik, M., Chain, C., George, R.,Vignola, F., 2002. A new operational model for satellite-derivedirradiances: description and validation. Sol. Energy 73, 307–317.http://dx.doi.org/10.1016/S0038-092X(02)00122-6.

Perez, R., Kivalov, S., Zelenka, A., Schlemmer, J., Hemker Jr., K., 2010.Improving the performance of Satellite-to-Irradiance models using thesatellite’s infrared sensors. In: Proceedings of: ASES Annual Confer-ence, Phoenix, Arizona, USA.

Perez, R., Schlemmer, J., Hemker, K., Kivalov, S., Kankiewicz, A.,Gueymard, C.A., 2015. Satellite-to-irradiance modeling – a newversion of the SUNY model. In: Proceedings of Proc. of 42nd IEEEPV Specialists Conference, New Orleans, LA.

Polo, J., Antonanzas-Torres, F., Vindel, J.M., Ramirez, L., 2014.Sensitivity of satellite-based methods for deriving solar radiation todifferent choice of aerosol input and models. Renewable Energy 68,785–792. http://dx.doi.org/10.1016/j.renene.2014.03.022.

Polo, J., Martin, L., Cony, M., 2012. Revision of ground albedoestimation in Heliosat scheme for deriving solar radiation fromSEVIRI HRV channel of Meteosat satellite. Sol. Energy 86, 275–282.http://dx.doi.org/10.1016/j.solener.2011.09.030.

Polo, J., Martın, L., Vindel, J.M., 2015. Correcting satellite derived DNIwith systematic and seasonal deviations: application to India. Renew-able Energy 80, 238–243. http://dx.doi.org/10.1016/j.renene.2015.02.031.

Polo, J., Zarzalejo, L.F., Ramirez, L., 2008. Solar radiation derived fromsatellite images. In: Badescu, Viorel (Ed.), Modeling Solar Radiationat the Earth Surface. Springer-Verlag (chapter 18).

Potter, C., Lew, D., McCaa, J., Cheng, S., Eichelberger, S., Grimit, E.,2008. Creating the dataset for the western wind and solar integrationstudy (U.S.A.). Wind Eng. 32, 325–338. http://dx.doi.org/10.1260/0309-524X.32.4.325.

Pozo-Vazquez, D., Wilbert, S., Gueymard, C.A., Alados-Arboledas, L.,Santos-Alamillos, F.J., Granados-Munoz, M.J., 2011. Interannualvariability of long time series of DNI and GHI at PSA, Spain. In:Proceedings of SolarPACES 2011 Conf., Granada, Spain.

Qu, Z., Gschwind, B., Lefevre, M., Wald, L., 2014. Improving HelioClim-3 estimates of surface solar irradiance using the McClear clear-skymodel and recent advances in atmosphere composition. Atmos. Meas.Tech. 7, 3927–3933. http://dx.doi.org/10.5194/amt-7-3927-2014.

Rigollier, C., Lefevre, M., Wald, L., 2004. The method Heliosat-2 forderiving shortwave solar radiation from satellite images. Sol. Energy77, 159–169. http://dx.doi.org/10.1016/j.solener.2004.04.017.

Ruiz-Arias, J.A., Dudhia, J., Gueymard, C.A., Pozo-Vazquez, D., 2013a.Assessment of the Level-3 MODIS daily aerosol optical depth in thecontext of surface solar radiation and numerical weather modeling.Atmos. Chem. Phys. 13, 675–692. http://dx.doi.org/10.5194/acp-13-675-2013.

Ruiz-Arias, J.A., Dudhia, J., Lara-Fanego, V., Pozo-Vazquez, D., 2013b.A geostatistical approach for producing daily Level-3 MODIS aerosoloptical depth analyses. Atmos. Environ. 79, 395–405. http://dx.doi.org/10.1016/j.atmosenv.2013.07.002.

Ruiz-Arias, J.A., Gueymard, C.A., Santos-Alamillos, F.J., Pozo-Vazquez,D., 2015a. Do spaceborne aerosol observations limit the accuracy ofmodeled surface solar irradiance? Geophys. Res. Lett. 42, 605–612.http://dx.doi.org/10.1002/2014GL062309.

Ruiz-Arias, J.A., Quesada-Ruiz, S., Fernandez, E.F., Gueymard, C.A.,2015b. Optimal combination of gridded and ground-observed solarradiation data for regional solar resource assessment. Sol. Energy 112,411–424. http://dx.doi.org/10.1016/j.solener.2014.12.011.

Schillings, C., Mannstein, H., Meyer, R., 2004. Operational method forderiving high resolution direct normal irradiance from satellite data.Sol. Energy 76, 475–484. http://dx.doi.org/10.1016/j.solener.2003.07.038.

Schumann, K., Beyer, H.G., Chhatbar, K., Meyer, R., 2011. Improvingsatellite-derived solar resource analysis with parallel ground-basedmeasurements. In: Proceedings of Proceeding from ISES Solar WorldCongress, Kasel, Germany.

Sengupta, M., Habte, A., Kurtz, S., Dobos, A., Wilbert, S., Lorenz, E.,Stoffel, T., Renne, D., Gueymard, C., et al., 2015. Best PracticesHandbook for the Collection and Use of Solar Resource Data forSolar Energy Applications. Golden, CO, NREL/TP-5D00-63112.

Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M.,Duda, M.G., Huang, X.-Y., Wang, W., Powers, J.G., 2008. 3Adescription of the Advanced Research WRF Version 3. NCARTechnical Note NCAR/TN-475+STR (Report).

Stoffel, T., Renne, D., Myers, D., Wilcox, S., Sengupta, M., George, R.,Turchi, C., 2010. Concentrating Solar Power. Best Practices Hand-book for the Collection and Use of Solar Resource Data. Nationalrenewable Energy Laboratory, Golden CO (Report).

Suri, M., Cebecauer, T., 2011. Requirements and standards for bankableDNI data products in CSP projects. In: Proceedings of: SolarPACES2011 Conf., Granada, Spain.

Suri, M., Cebecauer, T., 2014. Satellite-based Solar Resource data: Modelvalidation statistics versus user’s uncertainty. In: Proceedings of: Proc.of: ASES SOLAR 2014 Conference, San Francisco, 7–9 July 2014.

Thuman, C., Schitzer, M., Johnson, P., 2012. Quantifying the accuracy ofthe use of Measure–Correlate–Predict methodology for long-termsolar resource estimates. In: Proceedings of: American Solar EnergySociety, Denver CO.

Troccoli, A., 2015. Improving the accuracy of hourly satellite-derivedsolar irradiance by combining with dynamically-downscaled estimatesusing generalised additive models. Personal communication sent: 2015.

Vernay, C., Blanc, P., Pitaval, S., 2013. Characterizing measurementscampaigns for an innovative calibration approach of the globalhorizontal irradiation estimated by HelioClim-3. Renewable Energy57, 339–347. http://dx.doi.org/10.1016/j.renene.2013.01.049.

Vernay, C., Pitaval, S., Blanc, P., 2012. Calibration of long-term globalhorizontal irradiation estimated by HelioClim-3 through short-termlocal measurement campaigns: extending of the results to Europeanand African sites. In: Proceedings of: World Reneable Energy Forum,Denver.

Viana, T.S., Ruther, R., Martins, F.R., Pereira, E.B., 2011. Assessing thepotential of concentrating solar photovoltaic generation in Brazil withsatellite-derived direct normal irradiation. Sol. Energy 85, 486–495.http://dx.doi.org/10.1016/j.solener.2010.12.015.

Vignola, F., Grover, C., Lemon, N., McMahan, A., 2012. Building abankable solar radiation dataset. Sol. Energy 86, 2218–2229. http://dx.doi.org/10.1016/j.solener.2012.05.013.

Vindel, J.M., Polo, J., Antonanzas-Torres, F., 2013. Improving dailyoutput of global to direct solar irradiance models with groundmeasurements. J. Renewable Sustain. Energy 5, 063123.

Vuilleumier, L., Hauser, M., Felix, C., Vignola, F., Blanc, P., Kazantzidis,A., Calpini, B., 2014. Accuracy of ground surface broadbandshortwave radiation monitoring. J. Geophys. Res. 119, 13838–13860.http://dx.doi.org/10.1002/2014JD022335.

Wald, L., Ranchin, T., Lefevre, M., Albuisson, M., Remund, J., 2003.Increasing the spatial resolutionofgriddeddataby fusionwithotherdatasets. In: Proceedings of: Cairns, Australia, 8–11 July 2003, pp. 168–173.

Wilbert, S., Geuder, N., Schwandt, M., Krass, B., Jessen, W., Meyer, R.,Nouri, B., 2015a. Best Practices for Solar Irradiance Measurementswith Rotating Shadowband Irradiometers. International EnergyAgency. Solar Heating & Cooling Programme (Report).

Wilbert, S., Kleindiek, S., Nouri, B., Geuder, N., Habte, A., Schwandt,M., Vignola, F., 2015b. Uncertainty of Rotating ShadowbandIrradiometers and Si-Pyranometers Including the Spectral IrradianceError. In: Proceedings of SolarPACES 2015 Conference, Cape Town,South Africa.

Zarzalejo, L.F., Polo, J., Martın, L., Ramırez, L., Espinar, B., 2009. Anew statistical approach for deriving global solar radiation fromsatellite images. Sol. Energy 83, 480–484. http://dx.doi.org/10.1016/j.solener.2008.09.006.

Zhong, X., Kleissl, J., 2015. Clear sky irradiances using REST2 andMODIS. Sol. Energy 116, 144–164. http://dx.doi.org/10.1016/j.solener.2015.03.046.