Effects of Aerosols and Clouds on Solar Energy Production from Bifacial Solar Park in Kozani, NW Greece

Highlights

What are the main findings?

• First long-term analysis (20 years) of the effect of aerosols and clouds on solar energy production in a bifacial solar park in Eastern Europe. Aerosols dominate mainly in spring and summer while clouds in winter.
• Annual increase in solar energy production by +800.7 MWh, corresponding to an annual reduction of ~538 metric tons of CO₂ and a financial gain of ~12,827 €.

What is the implication of the main finding?

• Combining reanalysis and Earth Observation datasets (ERA5, CAMS and PVGIS) to quantify climatological drivers of solar energy production.
• The methodological framework, based on globally accessible datasets and standardized modeling procedures, is transferable and could be applied to other regions with bifacial photovoltaic infrastructures.

Abstract

The impact of aerosols and clouds on solar energy production is a critical factor for the performance of photovoltaic systems, particularly in regions with dynamic and seasonally variable atmospheric conditions. In Northwestern Greece, the bifacial solar park in Kozani—the largest in Eastern Europe—serves as a valuable case study for evaluating these effects over a 20-year period (2004–2024). By integrating ERA5 reanalysis data and CAMS satellite-based radiation products with modeling tools such as PVGIS, seasonal and annual trends in solar irradiance attenuation were investigated. Results indicate that aerosols have the greatest impact on solar energy production during spring and summer, primarily due to increased anthropogenic and natural emissions, while cloud cover exerts the strongest effect in winter, consistent with the region’s climatic characteristics. ERA5’s estimation of absolute energy output shows a strong correlation with CAMS satellite data (R² = 0.981), supporting its reliability for trend analysis and climatological studies related to solar potential dynamics in the Southern Balkans. The bifacial park demonstrates an increasing energy yield of approximately 800.71 MWh/year over the study period, corresponding to an annual reduction of ~538 metric tons of CO₂ and a financial gain of ~12,827 €. This is the first study in the Eastern Mediterranean that combined ERA5 and CAMS datasets with the PVGIS simulation tool in a long-term evaluation of bifacial PV systems. The combined use of reanalysis and satellite datasets, rarely applied in previous studies, highlights the importance of localized, climate-informed modeling for energy planning and management, especially in a region undergoing delignification and decarbonization.

1. Introduction

Climate change is widely recognized as the defining challenge of our era, with far-reaching implications not only for the environment but also for national economies and public policy frameworks [1,2]. In response, decarbonization and sustainability have emerged as central pillars in international climate agreements, with renewable energy sources taking a leading role in the “green energy” transition [3,4,5,6]. In this respect, solar energy plays an increasingly vital role in the reduction of fossil fuel consumption, enhancing environmental responsibility, and addressing global challenges such as climate change adaptation, energy poverty, and sustainable development [7,8,9]. The rise in the exploitation of solar energy brings to the fore the parallel rise of photovoltaic (PV) installations at smaller or larger scales [10]. Due to their rapid commercialization, PV systems have become the backbone of the renewable energy mix [11], especially in countries with high solar irradiance [12,13,14]. Despite their continuous rise, PV systems still face various limitations that affect their performance, which are directly related to climatic and meteorological conditions prevailing in each region [15,16]. In regions with high dust or pollution levels, electricity losses may exceed 60%, particularly in desert or semi-arid environments [17,18]. These impacts introduce uncertainty in energy forecasting and can complicate the integration of solar parks into the power grid. For this reason, localized studies focusing on atmospheric attenuation of solar radiation are essential for improving system design, operational efficiency, and policy planning.

Aerosols directly modify the Earth’s radiative balance and indirectly affect cloud microphysics, lifetime, and albedo (indirect effect) [19,20,21,22]. Key aerosol properties—such as particle number density, chemical composition, and size distribution—can alter cloud reflectivity and precipitation potential [23]. Katopodis et al. [24] reported that clouds significantly affect the downward solar radiation and decrease the incoming solar irradiance that reaches the PV panels, leading to losses in the final energy production. This result is directly related to the ability of clouds to block a part of the solar irradiance, decreasing mostly the direct radiation reaching the surface level, and enhancing the diffuse component [25,26]. The processes of reflection, absorption and scattering of solar radiation by clouds are highly uncertain due to macroscopic characteristics and microscopic physical properties that are involved in these processes such as the cloud type, the cloud amount, the cloud thickness and the size of cloud droplets, as well as the relative height between aerosol layers and clouds [27,28]. While clouds generally cause global attenuation of shortwave solar radiation, the multi-scattering and reflecting effects of clouds that do not obscure the sun’s disk, can lead to local radiation enhancement [29]. Because of their complex physico-chemical processes and their spatio-temporal variability, forecasting the clouds’ behavior is related to significant uncertainty. Tzoumanikas et al. [30] deployed an all-sky imaging system to study the Cloud Radiative Effect (CRE) over the city of Thessaloniki in Northern Greece during a 2-year period, revealing that CRE increased with cloud cover and decreased with the solar zenith angle. Moreover, Dalezios et al. [31] reported that cumuliform clouds are the primary precipitation-producing cloud types in Central and Northern Greece.

Bifacial photovoltaics (BPVs) represent a promising technological advancement in this context. By harvesting irradiance from both the front and rear sides of the panel, BPVs can increase total energy output, particularly in locations with high albedo and diffuse radiation. Vertically mounted BPVs have also shown advantages at higher latitudes (above 45°), where the solar altitude angle remains low. According to the 2020 International Technology Roadmap for Photovoltaics (ITRPV), the global market share of bifacial solar cells is expected to rise from 20% in 2020 to 70% by 2030 [32,33,34], driven by their superior performance. Jahangir et al. [35] focused on the analysis of the bifacial solar farm configurations, indicating that bifacial modules are expected to become the leading technology for the next-generation solar farms, especially in places like the Middle-East and South America. Huge bifacial solar parks are already operating around the world, including Europe, and innovative companies are leading the way in the construction of high-efficiency and low-cost bifacial photovoltaics [36,37,38,39].

The region of Western Macedonia, NW Greece, historically being a major lignite producer, is now transitioning toward a sustainable and low-carbon energy strategy in response to the environmental and public health concerns associated with lignite [40,41]. As the capital of Western Macedonia, Kozani city, was affected by intensive industrial activities during the last decades (lignin-fired power plants in the greater Kozani-Ptolemaida plain), which in combination with different emission sources related with urban and agricultural biomass burning activities resulted in substantial amounts of aerosol and air pollutants [42]. As the region phases out coal and seeks climate neutrality, understanding how atmospheric conditions influence solar energy output becomes critical for both technological and policy development.

According to King et al. [43], regardless of several other factors that may contribute to various degrees, the establishment of a global sustainable economy is possible only if there is a continuous and reliable source of energy to drive it. Several studies have analyzed the economic footprint of solar energy, as well as its impact on reducing Greenhouse Gas emissions (GHG), aligned with the Paris Agreement’s goals. In this respect, it is predicted that by 2050, the generation of solar energy will have increased to 48% due to ongoing economic and industrial expansion, while the installed capacity of PV technology has already increased from 40.334 in 2010 to 709.674 MW in 2020 and the solar module prices have fallen by up to 93% in the same period [44,45]. In addition, Güney [46] had collected data from 35 countries with various income levels for the period 2005–2018 and revealed that solar energy usage reduced CO₂ emissions in each case. Furthermore, solar panels consume 99% less water and produce 90% less pollutants than coal-based energy systems [47], since solar energy is an inexhaustible and carbon-free energy source worldwide [48]. Carvalho et al. [49] conducted research dealing with policy frameworks that are needed to deliver climate and ecological benefits from solar farms, concluding that solar farms can demonstrably achieve benefits for biodiversity and ecosystem services, if they are appropriately deployed and well-managed, while they contribute to Net Zero goals. Moreover, according to International Energy Agency (IEA) [50], clean energy investment increased nearly 50% from 2019 to 2023, reaching 1.8 trillion USD in 2023, with a growing rate of approximately 10% per year. In Greece, according to 2019 report of the Hellenic Association of Photovoltaic Companies (HELAPCO) [51], 7% of electricity demand is covered by PV, bringing the country in the third place worldwide with respect to PV contribution to electricity needs, and in the fifth place regarding the installed PV capacity per capita. Moreover, in 2023, Greece produced 19% of its electricity from photovoltaics, surpassing the Netherlands (18%), while it is ranked first in Europe in the contribution of solar energy to the total energy mix.

Moreover, International Energy Agency-IEA (2023) [52] presents a rigorous comparison of solar irradiance models—including those based on Coupled Model Intercomparison Project (CMIP6), ERA5 reanalysis, Copernicus Atmosphere Monitoring Service (CAMS), and satellite-derived datasets like Solargis and Photovoltaic Geographical Information System (PVGIS). This benchmark evaluated global horizontal irradiance (GHI) and direct normal irradiance (DNI) across 129 ground stations worldwide. It found that models using geostationary satellite imagery and reanalysis data—such as CAMS and PVGIS—consistently outperformed CMIP6-derived datasets in terms of bias and Root Mean Square Error (RMSE). For example, Solargis and PVGIS products based on the EUMETSAT Satellite Application Facility on Climate Monitoring (CM SAF) showed lower average deviation metrics, with RMSE values often below 20 W/m², while CMIP6-based models exhibited higher variability and less consistency across regions. Additionally, Copernicus Pan-European Climate Database (PECD) integrates ERA5 and CMIP6 data for energy-related variables and highlights how bias correction and downscaling improve CMIP6’s performance but still fall short of the accuracy achieved by reanalysis and satellite-based systems like PVGIS. Finally, CMIP6 models are primarily designed for long-term climate projections, not high-resolution solar resource assessment. Their limitations include coarse spatial resolution (often >100 km), limited temporal granularity (daily or monthly averages), indirect treatment of solar irradiance (inferred from cloud cover, humidity, etc.), and biases since without correction, CMIP6 outputs can deviate significantly from observed solar data. On the other hand, the combination of ERA5, CAMS and PVGIS has a benchmarking advantage in terms of high temporal (hourly) and spatial resolution (~30 km), assimilating satellite observations (Moderate Resolution Imaging Spectroradiometer-MODIS) for accuracy, with detailed aerosol optical depth and cloud properties, which are crucial for solar attenuation modeling includes both bias correction and uncertainty estimates for better representation of local terrain, aerosols, and cloud dynamics. Combining ERA5, CAMS, and PVGIS offers significantly improved accuracy over CMIP6 outputs for solar resource benchmarking. Studies show that PVGIS and CAMS-based models achieve mean bias errors (MBE) of ±3–5% and root mean square errors (RMSE) of 10–20 W/m², compared to CMIP6-derived datasets which often exhibit MBE of ±10–20% and RMSE of 30–50 W/m² [53,54].

Although several studies have examined the effects of aerosols and clouds on solar power generation [55,56,57,58], only a few have investigated the impact of these factors on bifacial power plants using long-term Earth Observation datasets and performance of modeling tools, while translating the results into financial benefits and emission reductions. This study addresses this scientific gap by investigating the interrelations between meteorological/climate factors and their direct and indirect effects on solar energy production at a newly established bifacial solar park in the Kozani region, NW Greece, the largest installation of its kind in Eastern Europe to date. Due to its favorable climatic conditions, geographic landscape, and high solar irradiance, the Kozani solar park is optimally positioned for efficient energy generation, capable of meeting a significant fraction of the annual electricity demand. This is the first study examining the impact of the two major atmospheric drivers, aerosols and clouds, on solar energy production in Western Macedonia, assessing their influence in terms of energy loss, economic implications and equivalent CO₂ emissions. A 20-year period (2004–2024) was selected to evaluate long-term trends using a combination of remote sensing, Earth Observation (EO) datasets and modeling techniques [59]. Furthermore, the analysis extends beyond solar radiation and energy performance to consider the socio-economic dimension, which is of equal importance for regional development, energy transition and policy planning. Translating technical findings into economic and environmental terms is essential for supporting evidence-based decisions and guiding long-term investments in solar energy infrastructure.

This paper is divided as follows: Section 2 describes the study area, while Section 3 presents the dataset and the applied methodology. Section 4 includes the main results, and finally, Section 5 summarizes the conclusions and future research in this area.

2. Study Area

In the hilly landscape of northern Greece, near Kozani city (the capital of Western Macedonia Prefecture), the largest bifacial solar power plant in Eastern Europe (Latitude: 40.4175°N, Longitude: 21.6557°E) has been in operation since April 2022 (see Figure 1). The facility spans a total area of 450 hectares and consists of 18 project sites. With a total capacity of about 204 MW, it hosts half a million bifacial solar panels that generate 320 million kWh of clean electricity annually. The solar power plant is situated on south-facing slopes offering optimal exposure to sunlight and maximizing energy output under clear-sky conditions.

Greece, as a Mediterranean country, exhibits a high sunshine duration throughout the year and especially during summer, making it ideally suited for solar energy applications [60,61]. Consequently, over the past few decades, investments in photovoltaic technologies have steadily increased, driven by the need for clean, affordable energy and alignment with national goals for decarbonization, delignification and energy transition. Figure 2 and Figure 3 show the rapid growth in the installed solar energy capacity in Greece over the last decade, as well as the rising share of employment in the photovoltaic sector relative to other renewable technologies. Notably, while solar capacity remained stable during the financial recession, it has grown rapidly after 2019. Employment in the photovoltaic sector now exceeds that of all other renewable energy sectors combined (Figure 3). Thus, solar development in Greece holds not only environmental significance—by reducing emissions from fossil fuel combustion—but also substantial socio-economic value.

Located at the intersection of air masses from Asia, Africa, and Europe, the Eastern Mediterranean—and Greece in particular—is frequently affected by diverse atmospheric conditions [63]. These air masses are composed of a complex mixture of natural dust aerosols from the Sahara and Middle East deserts, anthropogenic emissions from major regional megacities (i.e., Istanbul, Cairo, Athens), and smoke aerosols from seasonal forest fires [64]. More specifically, Greece receives high amounts of mineral dust from the Sahara Desert under certain meteorological conditions mainly in spring and summer [65,66], while smoke from local forest fires also plays a critical role in the concentration of light-absorbing soot particles during summer [67,68,69]. This aerosol mixture strongly affects solar radiation transmission, cloud microphysics, and atmospheric composition, leading to measurable reductions in photovoltaic efficiency and energy production [70,71].

In the Kozani region, cloudiness shows a significant inter-annual and intra-seasonal variability (Figure 4), which highly affects the amount of solar radiation reaching the surface, and consequently, the solar power generation. During the summer months (June to August), cloud cover remains minimal, with July showing peak clear-sky conditions—reaching up to 91% cloud-free days. On the contrary, from September to May, higher cloudiness is observed, peaking in December when clear-sky conditions drop to 49%. Therefore, according to clouds’ climatology, in Northern Greece, solar energy production is significantly impacted only in periods of heavy cloudiness (mainly between December and February). Moreover, during mid-end of January, winter bad weather days are often interrupted by clear sunny days, traditionally known as “Alkion days” [72,73].

3. Materials and Methods

3.1. Data

This study relies on simulated solar energy output data covering the period from 2004 to 2024. The surface solar radiation downwards (SSRD) and the top of the atmosphere (TOA) solar radiation data were obtained from ERA5, the new global reanalysis database provided from European Centre for Medium-Range Weather Forecasts (ECMWF), for the region of Western Macedonia, and more specifically, for the coordinates of the bifacial solar park. In addition, aerosol data over the same region were obtained from Copernicus Atmosphere Monitoring Service (CAMS). Moreover, for the evaluation of ERA5 reanalysis, solar irradiation data were also taken from CAMS radiation service through Solar Radiation and Data Services (SoDa Pro). For the determination of basic parameters, like system losses, and the calculation of the equivalent solar energy output, the Photovoltaic Geographical Information System (PVGIS) tool was also used.

The ERA5 provides hourly meteorological and climate data on single levels from 1940 to present with a spatial resolution of approximately 31 km (0.25° × 0.25°). Jiang et al. [74] focused on the evaluation of total, direct and diffuse irradiances from the ERA5 reanalysis data in China, indicating that the reanalysis outputs exhibited a good correlation with ground observations. More specifically, ERA5 overestimated hourly total solar radiation by 30.87 W/m² and the direct radiation by 73.95 W/m², while it underestimated the diffuse component by 43.08 W/m². In addition, intercomparison of four datasets namely Cloud, Albedo, Radiation dataset Edition 2 (CLARA), Surface Solar Radiation dataset—Heliosat Edition 2 (SARAH), ECMWF Reanalysis 5 (ERA5) and Arctic System Reanalysis v2 (ASR) for solar radiation analysis at high latitudes, revealed that ERA5 tends to overestimate the other datasets under clear skies and partly cloudy conditions, while it underestimated solar irradiance in overcast conditions [75].

Moreover, Khalil et al. [76] evaluated the performance of ERA5 solar radiation data in Helwan and Suez regions in Egypt during the period 2017–2019. The results revealed a strong correlation between ERA5 data and measured values (R = 0.95 and R = 0.97 in the two regions, respectively), highlighting that ERA5 is a reliable source for the evaluation of solar energy potential in Egypt and the Eastern Mediterranean. Additionally, comparison against pyranometer data in Automatic Weather Stations (AWS), showed that ERA5 solar radiation data exhibited a better accuracy (in comparison with NASA Power Dataset), with a coefficient of determination R² of 0.78, Root Mean Square Deviation (RMSD) equal to 199.42 W/m² and Mean Error (ME) of 245.99 W/m² [77]. Another study conducted in China from 2014 to 2020, dealing with ERA5 radiation product and its relationship with aerosols [78], indicated that the hourly total Surface Radiation (Rs) from ERA5 exhibited a better agreement with observations, while the hourly direct and diffuse radiation components (Rdir and Rdif) exhibited overestimation and underestimation, respectively.

Additionally, similar to [79], simulated solar radiation data were obtained through Solar Radiation and Meteorological Data Services (SoDa Pro) [80] and Photovoltaic Geographical Information System (PVGIS) tool (version 5.3) [81], for the collection of valuable data for reliable numerical computations. For the estimation process and the calculation of Aerosol Modification Factror (AMF) it was also important to use the Aerosol Optical Depth (AOD at 550 nm) for the same time period (2004–2024), which was sourced from CAMS. Vigkos et al. [82,83,84] presented a detailed analysis over 9 cities in Greece that supported the statistical robustness of the use of AOD 550 nm data for assessment of the aerosol effect on solar irradiance. The actual PV area, which is necessary for performing key calculations for the estimation of the equivalent total solar energy output, was determined through data provided by the park’s manufacturers and additional information obtained through PVGIS. As a result, it appears that the park covers a total area equal to 450 Ha and the actual PV area equals to 928.200 m² [85].

3.2. Methodology

This study focuses on the determination of the variation and seasonality of energy production in the bifacial solar park located in Kozani, along with the climatological trends in energy production (in GWh per year) and in influencing factors (aerosols and clouds). The trend results are translated into revenue and changes in the equivalent CO₂ emissions released.

ERA5 reanalysis was used for the extraction of SSRD and TOA solar radiation data, while the Global Horizontal Irradiance (GHI) data for all sky conditions were obtained from CAMS Radiation Service, which were used as a benchmark. The data comparison from both sources was crucial in order to assure their agreement and estimate their alignment.

Moreover, to investigate the impacts of aerosols and clouds on solar irradiance, two key indices were used: the Aerosol Modification Factor (AMF) and the Cloud Modification Factor (CMF), defined as follows:

$$\begin{gathered} AMF={\displaystyle \frac{{GHI}_0}{{GHI}_{00}}} \\ CMF={\displaystyle \frac{GHI}{{GHI}_0}} \end{gathered}$$

(1)

where GHI₀ indicates irradiance under clear-sky (cloud-free) conditions, GHI₀₀ represents irradiance under clean and clear-sky conditions (free of both aerosols and clouds), GHI refers to irradiance under all-sky conditions [82] and AOD refers to Aerosol Optical Depth at 550 nm. AMF and CMF express the relative reduction of solar radiation due to aerosols and clouds, respectively. Datasets were temporally harmonized to extract the influence of aerosols and clouds in solar radiation and to obtain their trends.

Additionally, for the conversion of solar irradiation data in energy output, the PVGIS application tool was used for the determination of the percentage conversion loss of the park and the actual PV area, as mentioned before. PVGIS examines a list of primary parameters to determine the system’s conversion efficiency such as the incidence angle, the PV technology and type, the azimuth and tilt of the system [86]. Through the synergy and combination of environmental and technical parameters, an integrated modeling of the complete energy chain was performed. The flowchart of the whole research initiative that took place in this study i.e., from data selection to solar-energy potential estimates, effects of aerosol and clouds and the assessment of emissions and revenue is shown in Figure 5.

To investigate the climatological trends, this study considers the long-term aerosol and cloud impacts on solar radiation by applying linear regression to high-resolution data during a 20-year period, with the resulting slope expressing the annual rate of change in units of Wh/m²/year. These trends signify persistent changes in atmospheric conditions (caused by anthropogenic activities or climate variability) that have an impact on solar radiation trends over the study area. A positive slope indicates an increasing influence of a particular factor that could affect solar energy production, whereas a negative slope indicates a diminishing impact. The identification of these trends, especially on a local scale, could provide valuable information about solar-energy production and reveal the key parameters in the decision-making process.

In general, solar irradiation, and hence, the subsequent solar energy production levels are a function of solar elevation, cloud microphysics, aerosol optical properties, water vapor, total ozone column etc. For the majority of affecting parameters monthly climatological values were used in order to bridge the gap between the input availability and the solar irradiation outputs accuracy. However, a preliminary investigation has been performed for the sensitivity of solar irradiation to water vapor and ozone. We compared integrated spectral GHI over the entire spectrum for different ozone values and we found a mean difference of only 0.5% for ozone ranging between 300 and 400 DU. For water vapor columns ranging between 0.5 and 2 cm we found a mean difference of 3.2%, although for solar elevation angles below 15 degrees this difference was higher, up to 5%. therefore, we chose neglect these variables in the first place and use total ozone column of 350 DU and water vapor of 0.5 cm for further calculations, considering differences mentioned above as a scale of error introduced for the purposes of this climatological study. To this direction, except the main two parameters, i.e., clouds and aerosols, we used monthly climatological values for the rest of the input parameters, more specifically ozone from OMI, water vapor from the Medium Resolution Imaging Spectrometer (MERIS) onboard ESA’s Environmental Satellite (ENVISAT), and angstrom and single scattering albedo from the AeroCom database [79,87].

Moreover, in this study we used surface solar irradiance measurements from ERA5 reanalysis and CAMS radiation service, both providing broadband shortwave radiation over the spectral range 285–2700 nm. This range contains the ultraviolet, visible, and near- to mid-infrared ranges of the solar spectrum relevant to photovoltaic applications, and includes absorption due to water vapour (H₂O) in the near- to mid-infrared. All our calculations throughout the analysis are made on the basis of this broadband spectral range, which forms the foundation for solar energy potential and PV performance assessment.

For the estimation of the Aerosol and Cloud Effects, which refer to the loss in radiation due to aerosol and cloud interactions in the atmosphere, respectively, the following equations were used:

$$\begin{gathered} Aerosol Effect= {GHI}_{00}-{GHI}_0 \\ Cloud Effect= {GHI}_0-GHI \end{gathered}$$

(2)

3.3. Energy, Emissions & Revenue Estimation

To quantify the impact of aerosols and clouds on energy losses at the PV park, the corresponding energy loss values were calculated as follows:

$$E_{loss,aerosols}=\left({GHI}_{00}-{GHI}_0\right)\times (1-n)\times ActualArea$$

(3)

for the energy loss due to aerosols,

$$E_{loss,clouds}=({GHI}_0-GHI)\times (1-n)\times ActualArea$$

(4)

for the energy loss due to clouds, and

$$E_{loss,total}=GHI\times (1-n)\times ActualArea$$

(5)

for the energy loss due to both aerosol and clouds.

Here, Actual Area refers to the actual PV area of the park and n is the conversions loss factor, both obtained from the PVGIS database. The “Actual Area” was considered constant over the 20-year period. This is a modeling choice, not to represent the real construction timeline of the Kozani PV park—which was completed in 2022—but to enable a consistent climatological simulation across the full 2004–2024 period. The above formulas account for the reduction in solar irradiance due to aerosols and clouds, adjusted for system efficiency and the effective area of solar panels. The linear trend of E_total was computed on a monthly basis for the actual PV area, in order to evaluate the impact of aerosols and clouds on energy performance. This resulted in an annual energy trend expressed in GWh/year, which is obtained by summing the trends of all months.

To assess the environmental impact of this energy loss, the corresponding amount of CO₂ emissions was estimated using the Greenhouse Gas (GHG) Equivalencies Calculator developed by the U.S. Environmental Protection Agency (EPA) [88]. This tool converts energy data into equivalent emissions based on standard combustion processes and presents the results in familiar environmental metrics, making them easier to interpret and apply in energy and policy planning.

Additionally, the economic impact of energy production was estimated by converting the total PV output into annual financial revenue as follows [89,90]:

$$Revenue\left(€\right)=Energy Production \left(\mathrm{k}\mathrm{W}\mathrm{h}\right)\times Electricity Cost(€/\mathrm{k}\mathrm{W}\mathrm{h})$$

(6)

where the electricity cost was calculated as the difference between the selling price (tariff) and the production cost per kWh:

$$Electricity Cost (€/\mathrm{k}\mathrm{W}\mathrm{h})=Selling Price (€/\mathrm{k}\mathrm{W}\mathrm{h})-Production Cost (€/\mathrm{k}\mathrm{W}\mathrm{h})$$

(7)

This approach allows for a combined assessment of environmental and economic outcomes associated with atmospheric influences on solar energy production, contributing to a more comprehensive understanding of the system’s performance and value.

4. Results and Discussion

4.1. ERA5 and CAMS Radiation Data

As mentioned above, validating the ERA5 reanalysis data against an independent source is essential for assessing its accuracy and reliability. For this purpose, GHI values obtained from CAMS were used as a benchmark. Both CAMS and ERA5 datasets provide surface solar radiation estimates under all-sky conditions, and their comparison is shown in Figure 6. This evaluation corresponds to daily data from 2004 to 2024 across the entire study area and targeting the coordinates of the bifacial solar park. The resulting scatter plot shows a strong linear correlation (R² = 0.981) indicating a high degree of agreement between the two datasets, thus supporting their use in climatological studies and long-term trend analysis, as any small, static biases do not affect the interannual variation and sign of the trends. The red linear regression line in Figure 6, highlights this close relationship, suggesting that CAMS and ERA5 solar radiation data are well-aligned and can be reliably used in parallel for applications such as solar energy modeling and climate studies.

Table 1. Characteristics of CAMS and ERA5 reanalysis datasets [50].

Data	ERA5	CAMS
Data Type	Reanalysis	Satellite
Spatial Resolution	31 km	Spatial interpolation at the location under study
Temporal Resolution	1 h	15 min
Radiative Transfer Model	RTTOVv11 [93]	LibRadtran [94]
Time Period	1940 to present	2004 to present
Area of Coverage	Global	Europe, Africa, Middle East, Eastern of South America and Atlantic Ocean

summarizes the main characteristics of the two datasets. The data that were used from CAMS Radiation Service have a 15 min temporal resolution over a 20-year period. For the estimation of GHI, CAMS uses the McClear model, while McCloud algorithm estimated the attenuation of radiation due to clouds. Both are implemented using the libRadtran radiative transfer model [91]. ERA5 provides data with 1-h temporal resolution over the same time frame (20 years) and a spatial resolution of 0.25° × 0.25° (~31 km). RTTOVv11 and “McRad” were used as the radiative transfer model and the radiation scheme, respectively. ERA5 SSRD values are originally provided on an hourly basis (in J/m²) and were converted to W/m² by dividing them by 3600 s [92]. Although both ERA5 and CAMS contain inherent uncertainties arising from differences in data modeling, spatial resolution, and retrieval methods, the high correlation coefficient indicates strong alignment of the data and their suitability for long-term solar energy analyses.

4.2. Aerosol and Cloud Effects

The evaluation of the effects of aerosols and clouds on the attenuation of solar energy production at the bifacial solar park is the main objective in this study. Using the methodology described above, the AMF and CMF time series were analyzed over the period 2004–2024 (Figure 7 and Figure 8, respectively). Both AMF and CMF are dimensionless coefficients ranging from 0 to 1. Values closer to 1 indicate minimal attenuation of solar radiation, while values closer to 0 reflect stronger attenuation, either due to increased aerosol loading or denser cloud cover.

Linear regression was applied to each time series, yielding the following trend equations:

$$CMF\left(x\right)=0.000001x+0.7773\hspace{1em}\hspace{1em}\&\hspace{1em}\hspace{1em}\hspace{1em}AMF\left(x\right)=0.000008x+0.7977$$

(8)

Both linear regressions exhibit positive slopes, indicating a gradual reduction in the influence of aerosols and clouds on the surface solar radiation over the 20-year period. The AMF displays a steeper and statistically significant slope (with statistical significance at 99% confidence level, a < 0.001) that suggests a more pronounced decline in aerosol influence compared to cloud-related attenuation during the last two decades. This trend is consisted with several studies conducted in Greece and the eastern Mediterranean, which agree to a declining aerosol tendency during the last decade, as well as to a decreasing trend in dust [95,96,97,98,99,100]. The observed increase in AMF suggests improved atmospheric transparency and, therefore, reduced solar radiation losses due to aerosols, supporting enhanced solar energy production in recent years. On the other hand, CMF exhibits greater seasonal variability and a broader range of values, reflecting the more dynamic and variable nature of cloud cover in modulating solar radiation over the study region.

To further explore the seasonal behavior of both factors, Figure 9 illustrates the monthly mean variability of AMF and CMF. AMF demonstrates a relatively moderate seasonal variation, with a decrease by the end of spring and during summer (e.g., ~0.77 in May). This decline corresponds to periods of elevated aerosol concentrations in Western Macedonia, often driven by seasonal wildfires [68], reduced precipitation, dry conditions, and the resuspension of dust from surrounding semi-arid landscapes. Additionally, Saharan dust intrusions, which occasionally reach northern Greece during these months [101], further reduce AMF values by increasing AOD, thereby enhancing attenuation of surface solar irradiance and lowering PV output. On the other hand, AMF peaks during winter (January December~0.91), signaling minimal aerosol influence. This seasonal peak reflects the generally lower aerosol loading in the atmospheric column over the southern Balkans during winter [102].

CMF, on the other hand, displays pronounced seasonality, with the highest values occurring in summer (July~0.875), when the sky over southern Europe and Greece is mostly cloud-free, and lowest values during the winter months (February~0.66), consistent with the region’s increased cloud cover during the colder months. Accordingly, it becomes evident that aerosols are the dominant factor in solar attenuation during spring and summer, while cloudiness is the primary limiting factor in winter, in line with the typical meteorological patterns observed in northern Greece [72,103]. Overall, the combined effects of clouds and aerosols revealed that the period from June to August, and especially July, is considered as the most favorable for solar power generation, as it’s associated with the lowest combined effect of these parameters. On the contrary, CMF and AMF exhibited their highest influence in winter (December to February) when cloudiness outweighs the reduced aerosol effect, resulting in an attenuation of solar irradiance.

4.3. Trend Analysis

Previous analysis showed that the seasonal behavior of AMF and CMF differs during the year, with aerosols exerting a stronger influence on solar radiation during the summer months, while clouds dominate attenuation during the winter. To complement this, we examined the trends in monthly downward surface solar radiation over the period 2004–2024. The slope values derived from linear regression analysis (in Wh/m²/year), shown in Figure 10, quantify the average interannual change in surface solar irradiance for each month. As shown in Figure 10, the SSRD trend values in the Kozani BPV solar park, range from approximately −534 Wh/m²/year in May to +554 Wh/m²/year in April. The sum of all the monthly trends leads to a total annual net trend of +1218.4 Wh/m²/year, indicating a clear improvement in the radiation-production conditions in the area over the last twenty years.

Moreover, Figure 11 illustrates the corresponding monthly trends in the climatological effects of aerosols and clouds (in Wh/m²/year) over the 20-year period. The integrated analysis reveals that the annual trend in radiation loss due to aerosols is −783.05 Wh/m²/year, while that due to clouds is −310.54 Wh/m²/year. In this context, a negative slope indicates a decreasing impact over time, while a positive slope would imply increasing attenuation. The negative trends observed for both components suggest that the combined effect of aerosols and clouds on solar attenuation has progressively weakened over the last two decades at the Kozani bifacial PV site.

The combined evaluation of the linear trends of SSRD and the climatological factors that influence it (aerosols and clouds) reveals distinct seasonal patterns and climatic variations. More specifically, significant increases of SSRD trends are observed in spring and autumn, with a peak in April (+554 Wh/m²/year) and October (+517 Wh/m²/year), while in May and August characteristic negative values of SSRD trends are presented (−534 Wh/m²/year and −395 Wh/m²/year, respectively). At the same time, in the peak positive SSRD trend values (April and October), a simultaneous downward trend in both radiation loss components is observed. More specifically, during the month of April, the Aerosol Effect trend is equal to −146 Wh/m²/year, while the Cloud Effect trend is equal to −109 Wh/m²/year. These negative trends indicate a reduction in radiation losses due to these two factors, a fact reflecting a cleaner atmosphere and less cloud cover during early spring over Kozani. In October, where SSRD shows the second highest positive peak, the Aerosol Effect trend is −101 Wh/m²/year and the Cloud Effect trend equals to −79 Wh/m²/year. This behavior is consistent with the autumn transition season, when less moisture and dust transport are usually observed. On the contrary, in the months with the most pronounced negative SSRD trends (May and August), the Cloud Effect trends show relatively small changes (May: −108 Wh/m²/year, August: −126 Wh/m²/year), while the Aerosol Effect was stronger (May: −260 Wh/m²/year, August: −196 Wh/m²/year). Therefore, the reduction trends in solar radiation in these months can be attributed mostly to aerosol rather than cloud variability.

Using Equation (5) and the methodology outlined above, with an assumed actual PV area of 928,200 m² and n = 21.15% system losses (as derived from PVGIS), these monthly trends were converted into actual energy production values. The results are summarized in

Table 2. Monthly trends of actual energy production values during the period 2004–2024 at Kozani’s bifacial solar park.

Month	Energy Production Trend (MWh/Year)
January	82.42
February	192.82
March	304.88
April	405.21
May	−390.87
June	−179.23
July	173.93
August	−179.80
September	186.05
October	343.45
November	−205.25
December	67.11

, showing the estimated change in energy production per month (in MWh/year). A positive trend (e.g., +82.42 MWh/year in January) reflects increased solar energy availability, while negative values denote reduced production. By aggregating the monthly energy trends, the annual net change in energy production was foreseen to increase by 800.71 MWh/year for the bifacial solar park. This value corresponds to a 95% confidence interval of ±159.63 MWh/year (approximately ±20%), which highlights the monthly and seasonal variability in the net annual increase in solar energy production. Moreover, this trend is calculated based on the assumption of a constant photovoltaic surface area (as mentioned in Section 3.3), which does not reflect the actual operating history of the park but is used exclusively as a climatological simplification.

In addition, based on data from both ERA5 and CAMS, the annual energy efficiency of the BPV park was calculated in GWh/year following the same methodology as above. The total annual solar radiation (for the year 2024, for example), obtained by summing the monthly values and based on data from ERA5, is calculated to be equal to 321.4 GWh/year, while based on CAMS data, it is calculated to be equal to 332.8 GWh/year. According to the manufacturer’s (juwi/HELPE Renewables) specifications, the declared theoretical annual energy output of the park is 320 GWh/year [85]. Therefore, the CAMS estimation overestimates the actual value by approximately 4%, while the ERA5 presents an overestimation of only 0.43%. These results validate the simulation values in the study area and strongly support their efficiency and reliability, providing the basis for the assessment of avoided CO₂ emissions and economic revenues presented in the next section.

4.4. Emissions & Revenue

After examining the overall solar energy trends and the dependence on aerosol and clouds variability, this section evaluates the energy trends in terms of avoided CO₂ emissions and financial revenue. Based on the technical characteristics of the bifacial solar park (which are available on the park’s website) and outputs from the PVGIS tool, the actual PV area (in m²), as well as the loss energy coefficient n were calculated.

According to publicly available financial data, the construction company operating the bifacial solar park in Kozani secured a fixed tariff of 56.72€ per MWh through a competitive auction process [104]. This value represents the selling price for the electricity generated. Furthermore, IRENA’s 2023 Renewable Energy Cost Report [105] indicates that the levelized cost of electricity (LCOE) for new utility-scale PV projects decreased by 12% between 2022 and 2023, placing the global average at US$0.044/kWh, or approximately €0.0407/kWh. Therefore, the net energy profit margin—defined as the difference between the selling price and the production cost—was estimated at €0.01602 per kWh (Equations (5) and (6) in Section 3.2 Methodology).

To translate energy into CO₂ emissions, as mentioned in methodology section, the Greenhouse Gas Equivalencies Calculator was used. To quantify avoided CO₂ emissions from reductions in electricity consumption (kWh), the Greenhouse Gas Equivalencies Calculator employs the marginal emission rates provided by the AVERT model [106], reflecting the U.S. national weighted average CO₂ emissions associated with displaced electricity generation. The current (from 2022) national marginal emission factor, as provided by Emissions & Generation Resource Integrated Database (eGRID), is 1405.3 lbs CO₂/MWh. After converting to metric units and accounting for transmission and distribution losses (~5.1%), this yields:

$$\begin{gathered} 1405.3 \mathrm{l}\mathrm{b}\mathrm{s} {\mathrm{C}\mathrm{O}}_2/\mathrm{M}\mathrm{W}\mathrm{h}\times 1 \mathrm{m}\mathrm{e}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c} \mathrm{t}\mathrm{o}\mathrm{n}/2204.6 \mathrm{l}\mathrm{b}\mathrm{s}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} \\ \begin{array}{}\times 1/\left(1-0.051\right) \mathrm{M}\mathrm{W}\mathrm{h} delivered/\mathrm{M}\mathrm{W}\mathrm{h} generated\\ \times 1 \mathrm{M}\mathrm{W}\mathrm{h}/1000 \mathrm{k}\mathrm{W}\mathrm{h}\\ =6.72\times {10}^{-4} \mathrm{m}\mathrm{e}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{c} \mathrm{t}\mathrm{o}\mathrm{n}\mathrm{s} {\mathrm{C}\mathrm{O}}_2/\mathrm{k}\mathrm{W}\mathrm{h}\end{array} \end{gathered}$$

(9)

All of the above estimates concerning revenue and CO₂ emissions, according to the energy trends that were calculated, are presented in

Table 3. Socio-economic parameter trend analysis for Kozani’s bifacial solar park.

Month	Energy Production Trend (MWh/Year)	Emissions CO2 (tons/Year)	Revenue (€/Year)
January	82.42	55.39	1320.37
February	192.82	129.58	3088.98
March	304.88	204.88	4884.18
April	405.21	272.3	6491.46
May	−390.87	−262.66	−6261.74
June	−179.23	−120.44	−2871.26
July	173.93	116.88	2786.36
August	−179.80	−120.83	−2880.4
September	186.05	125.03	2980.52
October	343.45	230.8	5502.07
November	−205.25	−137.93	−3288.1
December	67.11	45.1	1075.1

. The results can be interpreted in the following way: As an example, in January, a positive trend results in an increase in energy production of approximately 82.42 MWh/year. This additional yield leads to an estimated revenue of 1320.37 €/year and avoids around 55.39 metric tons of CO₂ emissions per year over the study period 2004–2024. In a second step, the summary of monthly solar energy production trends revealed a net annual increase of approximately 800.71 MWh/year. This enhancement in energy yield during the last two decades corresponds to an estimated reduction of 538.1 metric tons of CO₂ emissions per year, while the anticipated economic benefit amounts reached approximately 12,827.54 € annually. Applying the 95% confidence interval evaluator metric, the uncertainties (confidence intervals) in avoided emissions and revenues were estimated at ±107.27 tons of CO₂/year and ±2557.35 €/year, respectively. This revenue refers to the additional economic benefit resulting from the climatological trends and could be added to the total annual income of the solar park.

These findings highlight the climatic and economic feasibility of solar park deployment over time, reinforcing the potential of such projects in supporting decarbonization and long-term sustainable development goals, especially in the region of Western Macedonia, which hopes for a fair and sustainable energy transition. The progressive termination of the lignin-fired power plants in the Kozani-Eordaia Basin during the last decade has substantially reduced the aerosol emissions and concentrations, thus improving the atmospheric transparency to solar radiation reaching the ground, and in turn, the solar-energy production [107,108,109,110]. Further reduction in industrial emissions over the region would be beneficial for expanding of applications and investments on solar energy parks, also reflecting to economic benefits for the local society.

5. Conclusions

This study examined the critical impact of two key climatological factors, aerosols and clouds, on solar energy production in the largest bifacial photovoltaic solar park in Eastern Europe, located in Kozani region, NW Greece. By utilizing high-resolution satellite and reanalysis data over a 20-year (2004–2024) period, the study revealed seasonal and long-term climatological trends that affect solar energy production, as well as the equivalent CO₂ emissions and the economic returns. The strong correlation of R² = 0.981 between ERA5 and CAMS solar irradiance data supports their use in climatological studies and long-term trend analysis. The combined use of reanalysis and satellite data (like ERA5 and CAMS), integrated with modeling tools, provides a robust framework for monitoring and forecasting solar energy potential in dynamic atmospheric environments.

During spring and summer periods, aerosols were found to significantly reduce solar irradiance, while the dominant regulation factor during the winter season was clouds. The analysis identified an overall annual increase in energy production equal to 800.71 MWh/year during the study period, attributed to a reduction in aerosol loading and cloudiness over the region. This increasing trend in energy production is translated into a reduction of 538.1 metric tons of CO₂ emissions and a revenue of 12,827.54 € per year. These findings highlight the necessity of understanding the impact of meteorological and climatic factors like aerosols and clouds on energy strategy, both at the environmental level and in a socio-economic dimension.

Ultimately, this research reinforces the value of Earth Observation data, especially the combination of satellite and reanalysis data, a methodological approach that has not been extensively explored, as well as climatological trend analysis in solar energy planning and in sustainability strategies. Furthermore, the findings provide a new perspective in the decision making process, which is crucial for regions like the Western Macedonia, which is shifting from fossil fuels to a green energy future, aiming to attain a sustainable development.

This study presents a novel integration of ERA5, CAMS and PVGIS, and constitutes one of the first long-term (20 years) analyses combining reanalysis and Earth Observation datasets to quantify climatological drivers of solar energy performance in Southern Europe, offering a replicable methodology for other regions. The methodological framework developed here, which relies on globally accessible datasets (e.g., ERA5, CAMS, PVGIS) and standard modeling procedures, is transferable and could be applied to other regions with bifacial PV infrastructure, enabling climate-informed energy planning at regional or national scales. Depending on the environment, adaptations in the datasets may be required, for example higher CMF variability for humid tropical climatic zones. Future research in the study region may incorporate system-level operational factors such as power transmission losses, energy storage infrastructure, or socio-economic implications of the seasonal and annual variability of solar-energy production, thus focusing on societal-economic assessment of the Kozani’s solar park.

In December 2024, the first satellite of the Meteosat Third Generation (MTG) series, MTG-I1 (Imager), was put into operational service. It offers higher spatial resolution (down to 0.5 km) and temporal resolution (10 min for full-disk coverage and 2.5 min for rapid scanning of regions such as Europe), with 16 channels, which provide the ability to better detect clouds, aerosols and other meteorological features. In the future, the framework could assimilate next-generation cloud retrievals in near real-time to extract an observation-based Cloud Modification Factor and dynamically update the CAMS/ERA5 irradiance baseline via nowcasting, which would inform decision-making on a real time horizon.