Abstract

This study examines 2–3-day solar irradiance forecasts with respect to their application in solar energy industries, such as yield prediction for the integration of the strongly fluctuating solar energy into the electricity grid. During cloud-free situations, which are predominant in regions and time periods focused on by the solar energy industry, aerosols are the main atmospheric parameter that determines ground-level direct and global irradiances. Therefore, for an episode of 5 months in Europe the accuracy of forecasts of the aerosol optical depth at 550 nm (AOD550) based on particle forecasts of a chemical transport model [the European Air Pollution Dispersion (EURAD) CTM] are analyzed as a first step. It is shown that these aerosol forecasts underestimate ground-based AOD550 measurements by a mean of −0.11 (RMSE = 0.20). Using these aerosol forecasts together with other remote sensing data (ground albedo, ozone) and numerical weather prediction parameters (water vapor, clouds), a prototype for an irradiance forecasting system (Aerosol-based Forecasts of Solar Irradiance for Energy Applications, AFSOL) is set up. Based on the 5-month aerosol dataset, the results are then compared with forecasts of the ECMWF model and the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5), with Meteosat-7 satellite data, and with ground measurements. It is demonstrated that for clear-sky situations the AFSOL system significantly improves global irradiance and especially direct irradiance forecasts relative to ECMWF forecasts (bias reduction from −26% to +11%; RMSE reduction from 31% to 19% for direct irradiance). On the other hand, the study shows that for cloudy conditions the AFSOL forecasts can lead to significantly larger forecast errors. This also justifies an increased research effort on cloud parameterization schemes, which is a topic of ongoing research. One practical solution for solar energy power plant operators in the meanwhile is to combine the different irradiance models depending on the forecast cloud cover, which leads to significant reductions in bias for the overall period.

1. Introduction

Because of the limitation of fossil fuel resources and their impacts on climate change, our future energy system will increasingly depend on utilizing growing shares of renewable energy sources. This poses a major challenge to the development of future energy systems, since energy production from most renewable resources is highly variable in space and time. Because of this high variability, an efficient integration of solar energy into the existing energy supply system will only be possible if reliable information on ground-level solar irradiance is available. It should be noted that information on surface-level global irradiance is needed for photovoltaic facilities, whereas concentrating solar thermal power plants can only process direct irradiance.

To calculate direct and global irradiance at the surface level, exact information about clouds, aerosols, water vapor, and ozone is needed. For overcast skies, knowledge of cloud cover and type is most important in determining irradiance values. Cloud cover information is also important for distinguishing between overcast and clear situations. But in the clear-sky case, precise aerosol information is indispensable for providing accurate irradiance forecasts, since up to 20%–30% of direct irradiance extinction has been reported for cases of high particle occurrence (Henzing et al. 2004; Jacovides et al. 2000; Latha and Badarinath 2005). Despite the high spatial and temporal variabilities of aerosol occurrence, with typical scale lengths of a few hours or several tens of kilometers (Anderson et al. 2003; Holzer-Popp et al. 2008), most irradiance calculation systems use simple aerosol climatologies of fixed yearly or monthly values with a spatial resolution of several degrees instead of detailed aerosol information (e.g., Kinne et al. 2005; Schmidt et al. 2006).

A focus of the solar energy industry lies in relatively cloud-free regions, such as the Mediterranean area. This explains why clear-sky calculations are of great relevance for irradiance forecasts. The importance of accurate aerosol information is enhanced by the fact that these regions with comparably low cloud cover are often situated near desert regions, such as the Sahara Desert, which leads to frequent occurrences of dust transport and therefore episodically high atmospheric aerosol loads (Meloni et al. 2007).

So far, irradiance information has been primarily available as retrospective long time series used by the solar energy industry for site auditing and facility monitoring. For example, data from Meteosat satellites (Rigollier et al. 2004) is used to optimize solar power plants with respect to local characteristics. In addition, satellite-based data of the past 1–30 days are used to monitor the performance of existing solar energy sites, allowing for the near-real-time management of power sites as well as for retroactive performance checks by automatic failure detection routines (Drews et al. 2007).

However, near-real-time forecasts of direct and global solar irradiance are also needed for forecasts of facility yields of several hours to 2 days. Only with this information can sensible and efficient control and maintenance of solar power plants in combination with conventional plants be facilitated, as well as the stability of local and regional power grid systems. Additionally, besides air temperature, the amount of available solar irradiance largely determines customer consumption behavior. Therefore, irradiance forecasts are also needed when calculating consumer demands, an essential aspect in controlling both traditional and solar energy power plants.

Satellite-based cloud motion vectors for cloud fields and therefore surface irradiance are already used to calculate nowcasting irradiance predictions (Hammer et al. 1999). For example, the European Meteosat Second Generation (MSG) satellite provides high-resolution images of Europe and Africa every 15 min, where the movement of the clouds is used to determine and extrapolate motion vectors. Such an approach is of interest since it allows for forecast updates during the day, such as are needed for intraday adjustments of power plant management (Rikos et al. 2008) or energy trading applications. The accuracy of this method can be increased by the use of smoothing filter techniques (Lorenz 2004). However, it should be noted that this approach is limited to an approximate forecast horizon of up to 6 h.

One of the earliest approaches to predicting solar irradiance dealt with 1–2-day forecasts using model output statistics (MOS; Jensenius and Cotton 1981). The MOS technique uses statistical correlations between observed weather elements and climatological long-term data, satellite retrievals, or modeled parameters to obtain a highly localized statistical function. This allows, for example, for the adaptation of low-resolution mesoscale data to local conditions by considering local effects or by correcting systematic deviations of a numerical model, satellite retrievals, or climatological values. A disadvantage of this method is seen in the large amount of measurement data needed to develop statistical correlations separately for each location. This means that MOS-based forecasts are not available for larger areas or for locations without a priori information (Glahn and Lowry 1972).

For day-ahead production forecasts, global irradiance forecasts from numerical weather predictions such as from the European Centre for Medium-Range Weather Forecasts (ECMWF) are already operationally available for various grid sizes and on a global scale. However, temporal resolutions of 3 h as well as the restriction to global irradiance only instead of also direct irradiance needed for concentrating solar power systems, leads to problems when employing ECMWF irradiance forecasts for energy applications. Also, the forecasts are calculated with an aerosol climatology containing annual aerosol optical depth (AOD) cycles for four tropospheric aerosol types (ECMWF 2008) instead of actual aerosol forecasts, causing errors for clear-sky cases in particular.

For energy applications a combination of numerical weather prediction and air quality modeling is more suitable, where both meteorological and chemical forecasts rely on satellite- and ground-based data assimilation systems. The approach discussed in this paper is aimed at providing irradiance forecasts designed to meet the needs of the solar energy industries. This is achieved by improving the accuracy of clear-sky irradiance forecasts through the use of a chemistry transport model (CTM) for aerosol information, in combination with atmospheric input data from satellite and model data.

In section 2 the concept of an irradiance forecasting model is presented, followed by a detailed description of the atmospheric input parameters used. In section 3 the validation data sources are presented. The performance of the irradiance prediction system is detailed in section 4.

2. The AFSOL system

a. Concept

A modeling system of Aerosol-based Forecasts of Solar Irradiance for Energy Applications (AFSOL) has been developed to match the needs of the solar energy community regarding irradiance forecasts (Breitkreuz 2008). From this system, not only global but also direct irradiance information is available at high temporal resolution covering Europe and the Mediterranean region. A focus is placed on irradiance forecasts in clear-sky conditions since these include the most interesting situations for the efficient operation of solar energy plants.

All irradiance calculations of the AFSOL system are performed with the library for radiative transfer (libRadtran) program code (Mayer and Kylling 2005). The systems main routine, uvspec, calculates direct and global spectral irradiances at the surface level, taking into account atmospheric multiple scattering and absorption as well as surface properties. For this study the underlying standard atmosphere file is altered by various additional input data sources described in the following sections: aerosol information, column-integrated atmospheric water vapor, and cloud information is taken from a numerical weather prediction model, whereas atmospheric ozone content and ground albedo values are provided by satellite measurements (see Fig. 1).

Fig. 1.

AFSOL irradiance forecasting system scheme.

Fig. 1.

AFSOL irradiance forecasting system scheme.

The model is capable of calculating spectrally resolved irradiance values (Kato et al. 1999). However, due to the lack of sufficient numbers of appropriate validation measurements, most cases in this study consider only spectrally integrated global and direct irradiance predictions.

Various algorithms for solving the radiative transfer equations are available, which allows choosing according to different applications or required accuracies, such as between very exact time-consuming calculations or less exact faster routines applicable to operational services. For this study, the disort solver (Stamnes et al. 1988) is used for solar zenith angles of up to 70°, and the sdisort algorithm (Dahlback and Stamnes 1991) for solar zenith angles between 70° and 85°.

The temporal resolution of the direct and global irradiance forecasts is 1 h and the spatial resolution is a ½° grid across Europe and the Mediterranean African coastal regions. Each forecast is calculated for 72 h.

b. Atmospheric input data

Both cloud and aerosol information needed for the irradiance predictions are taken from the European Air Pollution Dispersion (EURAD) model (Ebel et al. 1997). This system has been developed for air quality monitoring and forecast purposes by the Rhenish Institute for Environmental Research at the University of Cologne, Cologne, Germany. It incorporates physical, chemical, and dynamical processes related to the emission, transport, and deposition of atmospheric substances. The system consists of three main submodels treating meteorological input from the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5; Grell et al. 1995), emission data (Memmesheimer et al. 1995), and a chemical transport model (Hass et al. 1995). An additional subsystem, the Modal Aerosol Dynamic Model (MADE) treats aerosol processes, which include particle emission, coagulation and growth, transport, and wet and dry deposition (Ackermann et al. 1998). Secondary organic species are accounted for within the Secondary Organic Aerosol Module (SORGAM) subsystem (Schell 2000).

The EURAD forecasts provide hourly data, with each forecast run covering 72 h. Depending on the target of the analysis, various grid sizes are available, ranging between a resolution of 1 km for local to regional studies and 125 km for hemispherical coverage. For the case study presented, a grid of approximately 54 km (½° grid) width was chosen, allowing coverage of Europe and the Mediterranean region from 30° to 60°N and from −10°W to 40°E with a reasonable amount of calculation resources.

1) Clouds and water vapor

Information on cloud parameters (cloud-top and -bottom height, cloud liquid water content, and cloud fraction) and total atmospheric water vapor is obtained from the meteorological part of the EURAD model. For information on cloud liquid water, vertically averaged values of all cloud-containing levels are calculated. This is a simplification of the three-dimensional EURAD cloud water output, which can be justified by the fact that a 54-km model grid is used: neglecting subgrid cloud structures masks any possible errors caused by the simplification of the vertical distribution of cloud liquid water.

The determination of the effective radius of cloud droplets is not part of the EURAD output. To calculate irradiances, a simple linear parameterization of radius growth with height is employed, consistent with the algorithm used at the ECMWF for shortwave irradiance calculations (ECMWF 2008).

2) Aerosols

The complete EURAD system yields mass concentrations of all treated species in three different size modes (nucleation, accumulation, and coarse), differentiated into 23 tropospheric height levels. Primary organic material and elemental carbon, sulfate, ammonium, nitrate, anthropogenic particulate matter, and aerosol liquid water are considered in the accumulation and nucleation modes. Anthropogenic aerosols are additionally included as coarse-mode particles.

A method of integrating natural coarse-mode particles (sea salt, fire particles, and dust) is in preparation and the effects of these additional modules will be subject to further investigations. This is a crucial point since, especially in the Mediterranean area, Saharan-based dust storms can occur frequently (Meloni et al. 2007), leading to a large extinction of direct irradiance. Consequently, the integration of external dust information into the aerosol model has a high priority from a solar energy point of view.

The particle mass concentrations of different aerosol species are usually combined to produce single PM10 values (total mass concentration of particles smaller than 10 μm at the surface, dedicated for air quality users) as standard output. In the case study presented, however, separate mass concentration values of all substances and size distributions modeled are used in order to calculate extinction coefficients for each grid point using a fast Mie extinction parameterization (Evans and Fournier 1990). Extinction coefficients σext are then vertically integrated through all height layers to produce aerosol optical depth values for each grid point.

In accordance with most studies regarding the influence of aerosols on solar radiation, AOD values presented in this paper are given at a wavelength of 550 nm (AOD550).

3) Ground albedo

As a source of ground albedo information data from the National Aeronautics and Space Administration’s (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua satellites (Schaaf et al. 2002) are used. Bimonthly composites of 1-km resolution are aggregated to match the ½° grid of EURAD aerosol output. To improve the retrieval accuracy, only pixels that were recorded in cloud-free atmospheres over land and with solar zenith angles of up to 60° are considered.

Satellite-based ground albedo data are difficult to validate since the measurement angles have to be the same in order to avoid anisotropy effects. However, validation studies over homogeneous semidesert terrain show that MODIS albedo values result in a mean RMSE of 0.02–0.07, thus deviating only slightly from ground measurements (Wang et al. 2004).

4) Ozone

For information on atmospheric ozone content, Total Ozone Mapping Spectrometer (TOMS) measurements from NASA’s Earth Probe Satellite are used (Bhartia 2004). The data (available online at http://wdc.dlr.de) offer global coverage and have a resolution of 1° × 1.25°. Accuracy studies show a deviation of TOMS total ozone columns against ground measurements of less than 2%, which is within the accuracy to be expected between different Dobson ground measurement devices (Bramstedt et al. 2002).

For this study daily mean values of ozone columns are used since the interdaily variability of ozone is not very high and because ozone’s influence is restricted to less than 1% when dealing with spectrally integrated irradiances (Mueller et al. 2004). In a few cases of limited data coverage, values from adjacent days are linearly interpolated.

c. Accuracy assessment of aerosol forecasts

The main purpose of this study is the evaluation of the coupling of an air quality model with numerical weather predictions to improve direct solar irradiance forecasts at the ground. Therefore, a validation of the aerosol model has to be conducted as a first step toward quantifying the aerosol input dataset accuracy and statistics. In a second step irradiance forecasts are validated against radiation measurements (section 3), in order to show the overall improvement in direct irradiance forecasting.

This validation study of the EURAD-based aerosol optical depth values was performed using Aerosol Robotic Network (AERONET) ground-based sun photometer measurements (Holben et al. 1998). The AERONET program is operated to gather aerosol information and provide validation data for satellite retrievals of aerosol optical properties. Datasets are available online (at http://aeronet.gsfc.nasa.gov) and contain AOD measurements at 16 different wavelengths at 1640, 1020, 870, 675, 667, 555, 551, 532, 531, 500, 490, 443, 440, 412, 380, and 340 nm, as well as solar zenith angles, total water vapor column measurements, and several variability coefficients used for automatic cloud screening procedures. The accuracy of AERONET AOD values is ±0.01 for wavelengths up to 440 nm and ±0.02 for shorter wavelengths (Holben et al. 1998).

In this study, all ground stations in Europe providing level 2 data (automatically cloud screened and manually inspected measurements) during a 5-month study period from July to November 2003 are considered. This leads to the use of 32 ground stations with a total of more than 70 000 measurements to be included in the analysis. Table 1 presents a complete list of all stations, including information about the number of measurements available, mean station AOD550, and corresponding variability information.

Table 1.

AERONET validation stations (in order of ascending latitude) with mean AOD550, standard deviation of AOD, and number of considered measurements.

AERONET validation stations (in order of ascending latitude) with mean AOD550, standard deviation of AOD, and number of considered measurements.
AERONET validation stations (in order of ascending latitude) with mean AOD550, standard deviation of AOD, and number of considered measurements.

For the whole time period and all locations considered, a mean underestimation by the modeling system of 0.11 (1σ = 0.16) was found for measured AOD550. This is not within the accuracy requirements aimed at for aerosol input data with regard to solar energy purposes, such as an RMSE of 0.10 (EHF and Ecole de Mines/Armines 2001). Thus, in the following several aspects of the accuracy analysis will be presented in order to identify crucial points for improvement.

The general underestimation is altered by a strong regional gradient, leading to good forecast accuracies in northern and middle Europe and significant underestimations in the central and western Mediterranean region as is clearly visible in the bias values of individual AERONET stations shown in Fig. 2. This pattern can be partly explained by the occurrence of forest fires in Portugal in August 2003, which is not considered in the model version used for the dataset analyzed in this paper. Also, some regions might be subject to misrepresentation in the emission database, leading to severe underestimations of the AOD, such as in the strongly industrialized Po Valley in northern Italy (Breitkreuz et al. 2007). The consideration of higher spatial resolutions is also expected to lead to reduced forecast errors in some of these cases, for example, for ground stations near large cities such as Palaiseau near Paris, France, or Munich, Germany, but is an approach that is neglected in this study for computational reasons, in order to allow the treatment of a larger study area.

Fig. 2.

Mean absolute bias (EURAD − AERONET) of AOD550 forecasts at the 32 ground measurement stations used; the sizes of the circles correspond to the AOD forecast variability.

Fig. 2.

Mean absolute bias (EURAD − AERONET) of AOD550 forecasts at the 32 ground measurement stations used; the sizes of the circles correspond to the AOD forecast variability.

However, a main reason lies in the occurrence of Saharan dust storms, which cannot be accounted for in the model system, thus leading to underestimations in AOD forecasts relative to ground measurements. The transport of Saharan dust across the Atlantic and the Mediterranean and toward middle Europe often lasts for 2–3 days and is typically caused by cyclones south of the Atlas Mountains, originating from the thermal contrast of cold maritime and warm continental air masses. In the central Mediterranean, most cases occur in summer, with a total occurrence of 4–5 days each month (Meloni et al. 2007). Consequently, the most severe underestimations of AOD550 can be found on days and in regions with Saharan dust outbreaks, as identified by visual analysis of MODIS or Advanced Very High Resolution Radiometer (AVHRR) satellite color composites (Breitkreuz 2008).

The mean forecast accuracy of single locations and also of all stations considered can be altered significantly by these short dust events. As demonstrated in Fig. 3, the daily mean values of absolute differences in AOD550 are strongly negative in cases of dust events. Dust events, such as on 21 July, were verified by visual analysis of color composite images from the MSG satellite and marked with black polygons in Fig. 3. As an example, when ignoring all measurements of the station of Etna, Sicily, during two 2-day dust episodes, where large particles were dominating the atmospheric aerosol load, the mean value of AOD difference for this location decreased from −0.16 AOD550 to −0.09 and the standard deviation for the whole time period changed from 0.23 to 0.14. This means that the integration of data on the atmospheric dust load has great potential to significantly improve the accuracy of aerosol forecasts from chemistry transport models. One approach, the assimilation of satellite-based aerosol measurements into the CTM, is currently being followed at the German Aerospace Center/German Remote Sensing Data Centre (DLR/DFD) together with the Rhenish Institute for Environmental Research at the University of Cologne, within the German Research Foundation’s (DFG) project Boundary Layer Aerosol Characterization from Space by Advanced Data Assimilation into a Tropospheric Chemistry Transport Model (AERO-SAM; Nieradzik and Elbern 2006; Martynenko et al. 2008).

Fig. 3.

Daily mean values of absolute differences in AOD550 (EURAD − AERONET, all stations). Days with marks coincide with dust events in the central Mediterranean as measured from ground stations and visually identified from MODIS color composites.

Fig. 3.

Daily mean values of absolute differences in AOD550 (EURAD − AERONET, all stations). Days with marks coincide with dust events in the central Mediterranean as measured from ground stations and visually identified from MODIS color composites.

Another significant pattern of forecast accuracy can be attributed to seasonal variations. Figure 3 already demonstrates that episodes with significant underestimation of AOD do occur in the summer months. Histograms of AOD error for separate months are shown in Fig. 4, where it becomes evident that during the summer months there is an absolute underestimation in AOD of −0.15 for July and August, whereas the aerosol predictions are much more accurate in September–November, with an underestimation of only −0.05.

Fig. 4.

Histogram of the absolute differences in AOD550 (EURAD − AERONET), separated for each month of the analysis period; normalized.

Fig. 4.

Histogram of the absolute differences in AOD550 (EURAD − AERONET), separated for each month of the analysis period; normalized.

This fact can partly be explained by the Saharan dust episodes in the western and central Mediterranean region, as mentioned above. Additionally, the summer of 2003 was exceptionally dry and dusty, with long periods without wet deposition. This caused an increase in atmospheric aerosol loads and therefore significantly increased AOD values over large areas of Europe (Hodzic et al. 2006). However, the model is not capable of reproducing these uncommonly long aerosol accumulation periods, which consequently led to underestimations in AOD during these summer months.

There is no interdependency between forecast errors and the time of day. A weak correlation between forecast length and accuracy can be established: the mean underestimation of AOD550 (EURAD model minus ground measurements) decreases from −0.13 (hours 1–24) to −0.11 (hours 25–48) to −0.09 (hours 49–72), while at the same time the standard deviation increases from 0.15 to 0.18. As a result, there is an apparently constant RMSE for all three forecast days, caused by the combination of these two reversed error tendencies.

Although there are still deficiencies in the EURAD-based AOD forecasting system, especially with regard to the treatment of desert dust particles, the model system is capable of reproducing the general features of the atmospheric aerosol load over Europe. Consequently, the approach of using aerosol information from a chemistry transport model for solar energy purposes seems promising—if the enhancements described above, such as the integration of dust information from near-real-time satellite data, will be pursued. It is shown in section 4 that using the EURAD AOD forecasts that have been assessed in section 2 yield a significant improvement in the irradiance forecasts despite their deficiencies.

3. Validation data sources

To quantify the accuracy of AFSOL irradiance forecasts, its results are compared to ground- and satellite-based measurements of global and direct irradiances for the period of July–November 2003. To enable comparisons with other available solar forecast data, the results are also checked against the performance of routinely available ECMWF and EURAD–MM5-based irradiance forecasts.

a. Ground measurements

For validation purposes ground measurements of global and (if applicable) direct irradiance from 121 locations in Europe and the Mediterranean area are used. Figure 5 shows their distribution throughout Europe. The accuracy of the pyranometers used at the various locations is given with a 2% deviation for the daily sum of the global irradiance, corresponding to the “secondary standard” of the World Meteorological Organization.

Fig. 5.

Locations of the 121 ground stations used for the validation of AFSOL irradiance forecasts.

Fig. 5.

Locations of the 121 ground stations used for the validation of AFSOL irradiance forecasts.

To improve compatibility, the sometimes hourly and sometimes instantaneous measurements of various temporal resolutions are all aggregated into hourly values, for easier comparison with the hourly AFSOL model values. However, it should be noted that the model data used are set on a rather coarse 54-km grid. This means that due to subgrid variabilities of clouds, and thus ground-level radiation, there will always be a certain minimal variability between hourly model values for an area (in this case, 2916 km2) and temporally averaged point measurements.

For 29 cases in which more than one ground measurement station can be assigned to an AFSOL grid box, an average RMSE of 16% for hourly global irradiance measurements and the complete 5-month period can be deduced (Breitkreuz 2008). This intrinsic variability of solar irradiance measurements, as the ground truth data, should be kept in mind when evaluating forecast accuracies: deviations of this order of magnitude are not necessarily due to incorrect forecasts, but might instead reflect natural subgrid variabilities of ground-level solar irradiance.

b. Other forecast data

Data from the ECMWF are used as a second source of irradiance forecasts. Solar global irradiance forecasts are operationally archived whereas direct irradiance is not available. To minimize the processing time, the computation of shortwave transmissivities is performed only every 3 h, using the values of temperature, specific humidity, liquid/ice water content, and cloud fraction at this time step, and climatologies for aerosols, atmospheric carbon dioxide, and ozone content (ECMWF 2008).

For this study atmospheric fields of the solar surface radiation downward (SSRD) parameter were obtained at a model grid of 0.5° × 0.5°. An arithmetic mean value of the four closest ECMWF grid points is attributed to each AFSOL grid point in order to balance out the somewhat incongruent grid systems. Each forecast starts at midnight. The radiation values are spectrally integrated from 200 to 4000 nm. All irradiation parameters at ECMWF are accumulated from the start of the forecast (in J m−2). To produce instant 3-h mean values for each time step given, irradiation values for each time step have to be isolated and normalized to the time interval. These 3-h values are then interpolated to obtain hourly mean values of irradiance. As linear interpolation leads to underestimations for high sun elevations and to overestimations for low sun, an interpolation method using the clear-sky index—the ratio of the forecasted global irradiance against the modeled clear-sky irradiance for the same situation (Girodo 2006)—is implemented here.

An additional model (Skartveit et al. 1998) is needed to separate the direct and diffuse components since the ECMWF only offers global irradiance values. Depending on the solar elevation and the clear-sky index, this allows for the determination of direct normal irradiance, which is the parameter needed for managing concentrating solar thermal plants. It has to be noted that there are slight distortions in the diurnal cycle of ECMWF global irradiances, caused by the fact that irradiance data need to be interpolated from 3-hourly values where no additional information on subhourly variability is available. These deviations tend to multiply when being transferred to the direct irradiance, leading to a slightly compressed daily curve of the ECMWF-derived direct irradiance values.

The accuracy of these temporally and spatially improved ECMWF forecasts has been analyzed for the summers 2003 and 2004 using data from 18 ground stations in Germany, where relative RMSEs of 14%–15% for cloud-free hourly values and of 35%–42% for all cloud situations are reported (Girodo 2006).

The meteorological part of the EURAD model also produces irradiance forecasts that are used as another source of comparison. These MM5-based data do not use aerosol climatologies to produce the global irradiance, like the operationally available ECMWF data. Instead, these data parameterize the clear-sky irradiance only as a function of solar zenith and climatological water vapor values (Grell et al. 1995). The MM5 global irradiance forecasts are employed in this study in order to show the importance of using accurate aerosol information instead of no explicit aerosol input, as in the EURAD–MM5 model.

c. Satellite-based measurements

Satellite-based irradiance measurements are used to give an impression of the theoretically possible accuracy when comparing spatially averaged data with point measurements. Global and direct irradiances from the European Meteosat-7 satellite, operated by the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), at a resolution of approximately 2.5 km × 4.5 km over middle Europe, are used for this study. For the retrieval of ground-level irradiance, the Heliosat algorithm (Hammer et al. 2007) is employed, which is used for Meteosat-7 and MSG operational retrievals at the Department of Energy and Semiconductor Physics of the Institute of Physics of the University of Oldenburg, Oldenburg, Germany.

The Heliosat method relies on the assumption that the radiance measured at the satellite, after ground and atmospheric reflection processes, is proportional to the atmospheric reflection (Cano et al. 1986) and thus mainly due to the impact of cloud reflection. Ground-level irradiance is then calculated from atmospheric transmission characteristics in combination with a model for clear-sky irradiance values. For the data used in this study, atmospheric aerosol input for the clear-sky model consists of a simple turbidity climatology (Dumortier 1998).

Together with geometrical corrections for the effects of cloud heights on the spatial distribution of ground-level irradiance as well as improvements in the determination of global irradiance in totally cloudy and cloud-free situations, this method has been attributed an RMSE of 19% and a bias of −0.5% for hourly values of global irradiance at 20 German locations for a period of 9 months during 2004 (Hammer et al. 2007).

4. Accuracy assessment of the AFSOL system

a. Clear-sky situations

For the period of July–November 2003, the performance of the AFSOL system is validated against ground-based measurements of 121 European sites described in section 3a (see Fig. 5). To allow for comparison with other available solar irradiance datasets, operationally available ECMWF global irradiance forecasts, EURAD–MM5-based global irradiance forecasts, and Meteosat-7 irradiance measurements are included.

Clear-sky cases are defined in this study by a low variability (σ < 0.02) of the clear-sky index based on the hourly ground measurements and a maximum cloud cover of 10% forecasted by both the EURAD–MM5 and the ECMWF model. This is a rather strict criterion employed to exclude any differences resulting from different cloud physical parameterizations in the ECMWF and MM5 models. Both models show different error statistics regarding clear-sky prediction, which shall not be discussed here in detail (Breitkreuz 2008).

For these clear-sky situations, AFSOL irradiance forecasts turn out to have a higher accuracy than operationally available ECMWF products. This is especially true for direct irradiance forecasts where the influence of aerosols is most relevant: the AFSOL system has a bias of +11.2% and an RMSE of 18.8%, whereas the ECMWF-derived direct irradiance forecasts underestimate by a mean of −26.3% (bias), together with an RMSE of 31.2%. Table 2 summarizes the statistical results of the intercomparison of the various irradiance datasets against ground-based measurements, whereas Fig. 6 shows histograms of their absolute deviations in global irradiance for cloud-free situations. As a comparison, the accuracy of direct irradiance measurements from Meteosat-7 leads to 15.6% (RMSE) and −1.7% (bias), thus giving the range of accuracy that can approximately be reached when comparing point measurements to 54-km-averaged model data.

Table 2.

Absolute and relative forecast accuracies (bias, RMSE) of direct irradiance forecasts (model − ground measurements).

Absolute and relative forecast accuracies (bias, RMSE) of direct irradiance forecasts (model − ground measurements).
Absolute and relative forecast accuracies (bias, RMSE) of direct irradiance forecasts (model − ground measurements).
Fig. 6.

Histogram of the absolute deviations in global irradiance for cloud-free situations in Europe: AFSOL, black; ECMWF, light gray; EURAD–MM5, dark gray; and Meteosat-7 retrieval, black dashed.

Fig. 6.

Histogram of the absolute deviations in global irradiance for cloud-free situations in Europe: AFSOL, black; ECMWF, light gray; EURAD–MM5, dark gray; and Meteosat-7 retrieval, black dashed.

This tendency is also valid for global irradiance forecasts: while the AFSOL forecasts can be characterized by a bias of +5.1% and an RMSE of 7.2%, the operationally available ECMWF forecasts lead to a bias of −9.8% and an RMSE of 11.5% (see Fig. 6). The EURAD irradiance forecasts have a bias of +5.0% and an RMSE of +9.4%.

On days with desert dust outbreaks, AFSOL direct irradiance forecasts in the central Mediterranean area have increased RMSE values relative to other regions. For example, if two 3-day-periods of dust outbreaks are eliminated from the whole 5-month period, the relative RMSE of all clear-sky situations decreases from 18.4% to 15.8% in the Mediterranean region, which then is within the accuracy range of other European regions such as southern Germany. This can be attributed to the fact that these regional dust episodes could not be modeled by the version of the EURAD system that is used for input in the aerosol forecasting system. Additionally, direct irradiance values are much more strongly influenced by extinction processes through dust particles than global irradiance. Consequently, this problem does not appear in global irradiance forecasts.

b. Cloudy situations

For cloudy situations the AFSOL system is less accurate than the ECMWF forecasting system, as is shown in Fig. 7, where the bias and RMSE of the global irradiance values of the different datasets for cloudy conditions are included. RMSE values for all model systems are closely coupled to the maximum cloud cover forecasts of the subset of situations included in the analysis (i.e., clear sky, maximum of 10% cloud cover, and maximum of 60% cloud cover; see Fig. 7). However, the accuracy of the AFSOL global and direct irradiance forecasts decreases more strongly when allowing cloud cover predictions of up to 100%, as compared to ECMWF data. This leads to relative RMSE values for global irradiance of up to 60% (AFSOL) for all cloud situations, or 37% (ECMWF) or 22% (Meteosat-7 measurements). It is pointed out, however, that the RMSE values of the EURAD–MM5 global irradiance forecasts are always higher than the RMSEs of the AFSOL predictions, showing the value of additional aerosol information.

Fig. 7.

Global irradiance forecast accuracy for different subsets of cloud situations in Europe, relative bias (dashed) and RMSE (solid): ECMWF, light gray; AFSOL, black; EURAD–MM5, black squares; and Meteosat-7 retrieval, dark gray.

Fig. 7.

Global irradiance forecast accuracy for different subsets of cloud situations in Europe, relative bias (dashed) and RMSE (solid): ECMWF, light gray; AFSOL, black; EURAD–MM5, black squares; and Meteosat-7 retrieval, dark gray.

While the mean underestimation of ECMWF global irradiance forecasts is at approximately −10%, regardless of the subset of cloud situations included in the analysis, the MM5-based irradiance systems (AFSOL and EURAD models) tend to significantly underestimate the results with increasing maximum cloud cover, causing a bias of up to −25% if all cloud cover situations are analyzed. This tendency can be explained by the effect that the MM5-based systems predict more high cloud-cover situations and less low cloud-cover situations relative to the ground-truth data.

For direct irradiance, similar tendencies are observed, with higher RMSE values for all subsets of cloud situations. This is shown in Fig. 8, where direct irradiance forecast biases and RMSEs of the different datasets are depicted for different subsets of cloud situations. Because of the more significant influence of aerosols on direct irradiance, the gain in accuracy in the clear-sky case from using actual aerosol forecasts instead of simple climatologies (AFSOL versus ECMWF) is more apparent here than for global irradiance (Figs. 7 and 8).

Fig. 8.

Direct irradiance forecast accuracies for different subsets of cloud situations in Europe, relative bias (dashed) and RMSE (solid): ECMWF, light gray; AFSOL, black; EURAD–MM5, black squares; and Meteosat-7 retrieval, dark gray.

Fig. 8.

Direct irradiance forecast accuracies for different subsets of cloud situations in Europe, relative bias (dashed) and RMSE (solid): ECMWF, light gray; AFSOL, black; EURAD–MM5, black squares; and Meteosat-7 retrieval, dark gray.

Forecast length has a significant impact on forecast accuracy, as long as cloudy situations are included in the analysis: for the AFSOL system, this can be quantified by RMSEs of 49.7% for the first day, 62.4% for the second day, and 67.7% for the third day. When considering only cloud-free cases, forecast length has no effect on bias or RMSE for any of the model systems analyzed. Thus, it can be deduced that this error tendency is caused exclusively by difficulties in cloud forecasts that increase with growing forecast duration.

The correlation of forecast accuracy with seasonal variation is only true for cloudy situations. Both direct- and global-irradiance forecasts of all model systems as well as the satellite measurements have smaller errors in the summer months (e.g., a relative RMSE of 55.9% for AFSOL global irradiance forecasts for July and August) than in the fall months (e.g., a relative RMSE of 68.9%). If only clear-sky situations are analyzed, no seasonal correlation at any location can be observed. This can be explained by the seasonal cycles in average cloud cover, leading to a higher percentage of cloudy situations and therefore larger errors in middle European autumn.

The different European regions also show a distinct pattern of forecast accuracy, when considering all cloud situations. RMSE values are highly correlated with average cloud cover for all considered models or retrieval systems. This leads to larger errors in the United Kingdom and around the Baltic Sea, whereas the Mediterranean region has higher forecast accuracies for both global- and direct-irradiance forecasts. Consistently, if only cloud-free situations are examined, no consistent regional tendency can be found.

c. Accuracy performance of combined forecasting systems

The two irradiance forecasting systems analyzed in this study have complementary strengths and weaknesses: while the AFSOL forecasts produce high errors for cloudy cases and very good agreements with ground measurements for clear-sky situations, the reverse is true for the ECMWF-based irradiance forecasts (see Figs. 7 and 8). This means that a possible practical way of improving overall irradiance forecasts is the combination of both models according to their strengths and thus eliminating the weaknesses of both approaches. This combined product will then have high forecast accuracies for clear-sky situations due to the AFSOL system, and acceptably high forecast accuracies for cloudy situations from the ECMWF data. Thus, it cannot only be used for direct irradiance applications, such as for concentrating solar systems, but also for global irradiance applications such as load forecasts.

It has to be clearly stated that the overall solution is an integration of an aerosol modeling scheme into the ECMWF model as currently prepared at ECMWF. But in the meantime, a practical solution that is easily implemented has to be found, as only this fulfills the daily operational needs, for example, of power plants that utilize concentrated solar energy.

If the AFSOL forecast is used for situations with forecasted EURAD–MM5 cloud cover of up to 10% and the ECMWF data for all other situations, then for the whole analysis period and all locations the bias for global irradiance can be reduced from −8% (ECMWF) and −25% (AFSOL) to −1% (combined method) with no changes in the RMSE (combined method and ECMWF: 37%). For the ground locations chosen and the period from July to November 2003, this is based on a ratio of approximately 20% AFSOL data and 80% ECMWF data. This means that combining both model systems is a promising practical way of improving the irradiance forecasts based on numerical weather prediction without an extended aerosol modeling scheme. However, the optimal cloud cover threshold value is subject to further investigation, especially with respect to the influence of different regions or seasons.

5. Discussion and conclusions

This study deals with short-term solar irradiance forecasts with respect to their application in solar energy industries. The main atmospheric parameter responsible for the extinction of solar irradiance is clouds. However, the main focus and economic potential of the solar energy industry is found in regions and time periods that have minimal cloud cover. During these “clear sky” cases, it is mainly aerosols—solid and liquid particles in the atmosphere—that influence the direct and diffuse irradiances at ground level. Aerosols are highly variable in space and time, which leads to difficulties in calculating and forecasting their spatiotemporal patterns and thus their influence on irradiance.

For an episode of 5 months (July–November 2003) in Europe, forecasts of the aerosol optical depth at 550 nm (AOD550) based on particle forecasts of the EURAD chemistry transport model are analyzed. It is shown that the aerosol forecasts underestimate ground-based measurements by a mean of −0.11 (RMSE of 0.20), which is not within the accuracy required by the solar energy industry for input parameters of irradiance forecasts. In particular, sporadic Saharan dust storm events in the central Mediterranean region lead to large inaccuracies that cannot be accounted for in the version of the EURAD model system used. Improvements in forecast accuracy can thus be expected from the integration of dust and also fire particle modules into the model system; these improvements are being prepared at the moment. Furthermore, the consideration of higher spatial resolutions is expected to lead to reduced forecast errors—an approach that has been neglected in this study for computational reasons, due to the spatial dimension of the study area. However, in other case studies the EURAD system has also been successfully applied on 15-, 5-, and 1-km scales.

Because of the high regional variability of aerosol presence and type, large differences in the representation accuracy for different European regions can be distinguished, such as severe underestimations of particle loads in the highly industrialized Po Valley in northern Italy or good results for remote continental areas in northern Europe. However, in spite of these deficiencies in the EURAD-based AOD forecasting system, especially with regard to the treatment of desert dust particles, the model system is capable of reproducing the general features of the atmospheric aerosol load over Europe.

Using these aerosol forecasts and other remote sensing data (ground albedo, ozone), as well as numerical weather prediction parameters (water vapor, clouds), a prototype for an irradiance forecasting system is set up: the Aerosol-based Forecasts of Solar Irradiance for Energy Applications system. Based on the 5-month dataset, its results are compared to forecasts of ECMWF and the MM5, satellite-based irradiance data from Meteosat-7, and ground measurements. It is demonstrated that for clear-sky situations the AFSOL system significantly improves direct irradiance forecasts compared to ECMWF forecasts, with a reduction in the relative bias from −26% to +11% and a reduction in the relative RMSE from 31% to 19%.

On days with desert dust outbreaks, AFSOL direct irradiance forecasts in the central Mediterranean area have increased RMSE values relative to other regions. This means that the integration of dust information into the model system, such as through the assimilation of satellite-based data in the forecasting system, would significantly improve the direct irradiance forecast accuracy for clear-sky situations especially in the Mediterranean region, both of which are areas of interest for the solar energy industries.

Global irradiance forecasts in the clear-sky case are also shown to have higher accuracies in comparison to the operationally available ECMWF forecasts, with a reduction in the relative bias from −10% to +5% and a reduction in the relative RMSE from 12% to 7%. However, for cloudy situations, the AFSOL forecasts can lead to significantly larger forecast errors due to cloud modeling deficiencies in the underlying mesoscale numerical weather model. Also, it should be pointed out that for all cloud situations except completely cloud-free cases the accuracy of MM5 irradiance forecasts is lower than for the ECMWF predictions.

This means that both direct and global irradiance forecasts of the AFSOL model show higher accuracies than ECMWF global irradiance forecasts or the derived direct irradiance products in clear-sky cases. Since in cloud-free situations the aerosols are the dominant parameter for determining solar irradiance at ground level, we suggest that the enhanced aerosols’ input into the irradiance calculation schemes significantly contributes to this accuracy improvement. This is valid even if there are still deficiencies within the aerosol forecasting system, especially regarding the representation of dust episodes, because the enhanced aerosol information still provides an improvement relative to the AOD climatologies used operationally at ECMWF.

In conclusion, regarding the application of the AFSOL system for solar energy purposes, it can be stated that for cloudy situations the AFSOL model in its current state is not well suited to forecasting irradiance and power loads: a forecast of consumer behavior and thus power consumption is only possible when cloudy and cloud-free situations can be forecasted accurately enough. However, the AFSOL systems produces good agreement with ground measurements for cloud-free situations and especially for direct irradiance forecasts—both of which determine the situations of greatest relevance for the management of solar thermal and photovoltaic power plants. Therefore, the need for an inclusion of a more detailed aerosol scheme into the ECMWF operational model has been clearly shown in this study. In the meantime, the combined use of ECMWF and AFSOL irradiance forecasts, depending on the level of forecasted cloud cover, can be recommended to fulfill daily operative needs at present-day concentrating solar power plants. This approach has been tested with satisfying results: a reduction in bias from −25% (AFSOL) and −8% (ECMWF) to −1% (combination method) is achieved.

Possible benefits from using the AFSOL model for optimizing the management strategies of a concentrating solar thermal power plant in Spain are described in a separate case study (Wittmann et al. 2008). Further test cases and the use of larger datasets for the development of power plant operation strategies are foreseen.

Acknowledgments

We thank the Rhenish Institute for Environmental Research of the University of Cologne, especially Lars Nieradzik, for providing large amounts of EURAD-CTM aerosol and EURAD/MM5 meteorological data. As well, we thank the AERONET PIs and their staff for establishing and maintaining the 32 sites used in this investigation.

For the provision of ground-based irradiance measurements at 121 sites, we thank the ECMWF, the Deutsche Wetterdienst (DWD), the Spanish Instituto Nacional de Meteorología, the Met Office, the AERONET network, the National Observatory Athens, the Global Atmospheric Watch Programme, the International Daylight Measurement Project, the Sveriges Meteorologiska och Hydrologiska Institut, the CEOP project, the Baseline Surface Radiation Network, the Institute of Construction and Architecture of the Slovak Academy of Sciences, the Solar Millennium AG, the Centre Universitaire d’Étude des Problèmes de l’Énergie of the University of Geneva, and the colleagues at the Plataforma Solar de Almería.

Our acknowledgment also goes to the developers of libRadtran (information online at www.libradtran.org) for their radiative transfer tools, which were used for all irradiance calculations.

We thank Dr. Elke Lorenz from the Energy and Semiconductor Research Department of the University of Oldenburg for Meteosat-7 retrieval data and for routines and help with the ECMWF data interpolation procedures.

This work was financed by the Virtual Institute of Energy Meteorology (vIEM), supported by the “Impuls- und Vernetzungsfond” of the Helmholtz Foundation.

REFERENCES

REFERENCES
Ackermann
,
I. J.
, and
Coauthors
,
1998
:
Modal Aerosol Dynamics Model for Europe: Development and first applications.
Atmos. Environ.
,
32
,
2981
2999
.
Anderson
,
T. L.
, and
Coauthors
,
2003
:
Mesoscale variations of tropospheric aerosols.
J. Atmos. Sci.
,
60
,
119
136
.
Bhartia
,
M.
,
2004
:
Algorithm theoretical basis document.
Version 8: TOMS total ozone algorithm. NASA Goddard Space Flight Center, MD, 23 pp. [Available online at http://toms.gsfc.nasa.gov/version8/v8toms_atbd.pdf]
.
Bramstedt
,
K.
, and
Coauthors
,
2002
:
Comparison of total ozone from the satellite instruments GOME and TOMS with measurements from the Dobson network 1996–2000.
Atmos. Chem. Phys. Discuss.
,
2
,
1131
1157
.
Breitkreuz
,
H.
,
2008
:
Solare Strahlungsvorhersagen für energiewirtschaftliche Anwendungen—Der Einfluss von Aerosolen auf das solare Strahlungsangebot. (Solar irradiance forecasts for energy applications—The influence of aerosols on solar irradiance.) Ph.D. thesis, Universität Würzburg, Würzburg, Germany, 153 pp
.
Breitkreuz
,
H.
, and
Coauthors
,
2007
:
A case study to prepare for the utilization of aerosol forecasts in solar energy industries.
Sol. Energy
,
81
,
1377
1385
.
Cano
,
D.
, and
Coauthors
,
1986
:
A method for the determination of the global solar radiation from meteorological satellite data.
Sol. Energy
,
37
,
31
39
.
Dahlback
,
A.
, and
K.
Stamnes
,
1991
:
A new spherical model for computing the radiation field available for photolysis and heating at twilight.
Planet. Space Sci.
,
39
,
671
683
.
Drews
,
A.
, and
Coauthors
,
2007
:
Monitoring and remote failure detection of grid-connected PV systems based on satellite observations.
Sol. Energy
,
81
,
548
564
.
Dumortier
,
D.
,
1998
:
The Satellight model of turbidity variations in Europe.
Sixth SATELLIGHT Meeting Rep., Ecole Nationale des Travaux Publics de l’Etat Vaulx-en-Velin, France, 11 pp
.
Ebel
,
A.
, and
Coauthors
,
1997
:
Air pollution studies within the EURAD Model System (3): EURAD—European Air Pollution Dispersion Model System.
Institute Geophysics and Meteorology Rep. Tech Rep. 6/1997, University of Cologne, Cologne, Germany, 134 pp
.
ECMWF
,
cited
.
2008
:
2.5: Input to the radiation scheme.
Integrated Forecast System documentation CY28r1. [Available online at http://www.ecmwf.int/research/ifsdocs/CY28r1/Physics/Physics-03-6.html]
.
EHF and Ecole de Mines/Armines
,
2001
:
Compilation of data requirements.
Heliosat-3 project, Contract NNE5-2000-00413, Fifth EU Framework Programme, 14 pp. [Available online at http://www.heliosat3.de/documents/Heliosat3_D2_data_req.pdf]
.
Evans
,
B. T. N.
, and
G. R.
Fournier
,
1990
:
Simple approximation to extinction efficiency valid over all size parameters.
Appl. Opt.
,
29
,
4666
4670
.
Girodo
,
M.
,
2006
:
Solarstrahlungsvorhersage auf der Basis numerischer Wettermodelle. (Solar irradiance forecasts based on numerical weather prediction models.) Ph.D. thesis, University of Oldenburg, Oldenburg, Germany, 159 pp
.
Glahn
,
H.
, and
D.
Lowry
,
1972
:
The use of model output statistics (MOS) in objective weather forecasting.
J. Appl. Meteor.
,
11
,
1203
1211
.
Grell
,
G. A.
,
J.
Dudhia
, and
D. R.
Stauffer
,
1995
:
A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5).
NCAR Technical Note NCAR/TN-398+STR, 138 pp
.
Hammer
,
A.
, and
Coauthors
,
1999
:
Short-term forecasting of solar radiation: A statistical approach using satellite data.
Sol. Energy
,
67
,
139
150
.
Hammer
,
A.
,
E.
Lorenz
, and
S.
Petrak
,
2007
:
Fernerkundung der Solarstrahlung für Anwendungen in der Energietechnik.
(Remote sensing of solar irradiance for applications in energy technologies.) Proc. Deutsch–Österreichisch–Schweizerische Meteorologentagung 2007, Hamburg, Germany, German–Austrian–Swiss Meteorological Association. [Available online at http://meetings.copernicus.org/dach2007/download/DACH2007_A_00357.pdf]
.
Hass
,
H.
,
H. J.
Jakobs
, and
M.
Memmesheimer
,
1995
:
Analysis of a regional model (EURAD) near surface gas concentration predictions using observations from networks.
Meteor. Atmos. Phys.
,
57
,
173
200
.
Henzing
,
J. S.
, and
Coauthors
,
2004
:
Effect of aerosols on the downward shortwave irradiances at the surface: Measurements versus calculations with MODTRAN4.1.
J. Geophys. Res.
,
109
,
D14204
.
doi:10.1029/2003JD004142
.
Hodzic
,
A.
, and
Coauthors
,
2006
:
Evolution of aerosol optical thickness over Europe during the August 2003 heat wave as seen from CHIMERE model simulations and POLDER data.
Atmos. Chem. Phys.
,
6
,
1853
1864
.
Holben
,
B. N.
, and
Coauthors
,
1998
:
AERONET—A federated instrument network and data archive for aerosol characterization.
Remote Sens. Environ.
,
66
,
1
16
.
Holzer-Popp
,
T.
, and
Coauthors
,
2008
:
Benefits and limitations of the new version synergetic aerosol retrieval SYNAER.
Satellite Aerosol Remote Sensing over Land, A. A. Kokhanovsky and G. de Leeuw, Eds., Springer, 227–266
.
Jacovides
,
C. P.
, and
Coauthors
,
2000
:
Spectral solar irradiance and some optical properties for various polluted atmospheres.
Sol. Energy
,
69
,
215
227
.
Jensenius
,
J.
, and
G. F.
Cotton
,
1981
:
The development and testing of automated solar energy forecasts based on the model output statistics (MOS) technique.
Proc. First Workshop on Terrestrial Solar Resource Forecasting and on the Use of Satellites for Terrestrial Solar Resource Assessment, Washington, DC, American Solar Energy Society, 22–29
.
Kato
,
S.
, and
Coauthors
,
1999
:
The k-distribution method and correlated-k approximation for a shortwave radiative transfer model.
J. Quant. Spectrosc. Radiat. Transf.
,
62
,
109
121
.
Kinne
,
S.
, and
Coauthors
,
2005
:
An AeroCom initial assessment—Optical properties in aerosol component modules of global models.
Atmos. Chem. Phys. Discuss.
,
5
,
8285
8330
.
Latha
,
K. M.
, and
K. V. S.
Badarinath
,
2005
:
Spectral solar attenuation due to aerosol loading over an urban area in India.
Atmos. Res.
,
75
,
257
266
.
Lorenz
,
E.
,
2004
:
Methoden zur Beschreibung der Wolkenentwicklung in Satellitenbildern und ihre Anwendung zur Solarstrahlungsvorhersage (Methods for the description of cloud development in satellite images and their application for solar irradiance forecasts). Ph.D. thesis, University of Oldenburg, Oldenburg, Germany, 111 pp
.
Martynenko
,
D.
, and
Coauthors
,
2008
:
Sensitivity analysis of synergetic aerosol retrieval and its application in assimilation procedure.
Proc. European Geophysical Union General Assembly, Vienna, Austria, European Geophysical Union, EGU2008-A-04030
.
Mayer
,
B.
, and
A.
Kylling
,
2005
:
The libRadtran software package for radiative transfer calculations: Description and examples of use.
Atmos. Chem. Phys.
,
5
,
1855
1877
.
Meloni
,
D.
, and
Coauthors
,
2007
:
Seasonal behavior of Saharan dust events at the Mediterranean island of Lampedusa in the period 1999–2005.
Atmos. Environ.
,
41
,
3041
3056
.
Memmesheimer
,
M.
, and
Coauthors
,
1995
:
Modeling of episodic emission data for Europe with the EURAD emission model EEM.
Proc. Int. Specialty Conf. on Regional Photochemical Measurement and Modeling Studies, San Diego, CA, Air and Waste Management Association, 495–499
.
Mueller
,
R. W.
, and
Coauthors
,
2004
:
Rethinking satellite-based solar irradiance modeling: The SOLIS clear sky module.
Remote Sens. Environ.
,
91
,
160
174
.
Nieradzik
,
L.
, and
H.
Elbern
,
2006
:
Variational assimilation of combined satellite retrieved and in situ aerosol data in an advanced chemistry transport model.
Proc. ESA Atmospheric Science Conf., Frascati, Italy, European Space Agency, 2006-440
.
Rigollier
,
C.
,
M.
Lefèvre
, and
L.
Wald
,
2004
:
The method Heliosat-2 for deriving shortwave solar radiation from satellite images.
Sol. Energy
,
77
,
159
169
.
Rikos
,
E.
,
S.
Tselepis
,
C.
Hoyer-Klick
, and
M.
Schroedter-Homscheidt
,
2008
:
Stability and power quality issues in microgrids under weather disturbances: An implementation to the Kythnos island power system.
IEEE J. Special Topics Earth Observations Remote Sensing
,
3
,
170
179
.
Schaaf
,
C. B.
, and
Coauthors
,
2002
:
First operational BRDF, Albedo Nadir Reflectance Products from MODIS.
Remote Sens. Environ.
,
83
,
135
148
.
Schell
,
B.
,
2000
:
Die Behandlung sekundärer organischer Aerosole in einem komplexen Chemie-Transport-Modell (The treatment of secondary organic aerosols in a complex chemistry transport model).
Institute for Geophysics and Meteorology Rep. 135, University of Cologne, Cologne, Germany, 171 pp
.
Schmidt
,
G. A.
, and
Coauthors
,
2006
:
Present-day atmospheric simulations using GISS ModelE: Comparison to in situ, satellite, and reanalysis data.
J. Climate
,
19
,
153
192
.
Skartveit
,
A.
,
J. A.
Olseth
, and
M. E.
Tuft
,
1998
:
An hourly diffuse fraction model with correction for variability and surface albedo.
Sol. Energy
,
63
,
173
183
.
Stamnes
,
K.
, and
Coauthors
,
1988
:
Numerically stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media.
Appl. Opt.
,
27
,
2502
2509
.
Wang
,
K.
,
J.
Liu
,
X.
Zhou
,
M.
Sparrow
,
M.
Ma
,
Z.
Sun
, and
W.
Jiang
,
2004
:
Validation of the MODIS global land surface albedo product using ground measurements in a semidesert region on the Tibetan Plateau.
J. Geophys. Res.
,
109
,
D05107
.
doi:10.1029/2003JD004229
.
Wittmann
,
M.
, and
Coauthors
,
2008
:
Case-studies on the use of solar irradiance forecasts for optimized operation strategies of solar thermal power plants.
IEEE J. Special Topics Earth Observations Remote Sensing
,
1
,
18
27
.
doi:10.1109/JSTARS.2008.2001152
.

Footnotes

* Current affiliation: Stadtwerke München GmbH, Munich, Germany.

Corresponding author address: Hanne Breitkreuz, German Aerospace Center, German Remote Sensing Data Center, Postfach 1116, 82234 Wessling, Germany. Email: hanne.breitkreuz@gmx.de