2.1

CyCARE campaign

The CyCARE (Cyprus Cloud Aerosol and Radiation Experiment) campaign represents a significant collaborative effort between the Cyprus University of Technology (CUT) and the Leibniz Institute for Tropospheric Research (TROPOS), aimed at advancing our understanding of aerosol-cloud dynamics and interactions under the unique atmospheric conditions prevalent in the East Mediterranean, north Africa and the Middle East (EMMENA region). Conducted in Limassol, Cyprus (34.675° N, 33.04° E, 22 m asl), the campaign spanned from October 2016 to April 2018. This initiative was particularly focused on examining the dynamics of dust and aerosol pollution and their influence on cloud formation and properties— an area of research critical for elucidating the complex feedback mechanisms affecting regional and global climate patterns.

A key aspect of the CyCARE campaign was the deployment of the Leipzig Aerosol and Cloud Remote Observations System (LACROS) at the premises of CUT. The LACROS setup is a comprehensive suite of both active and passive remote-sensing instruments designed to capture a multidimensional picture of atmospheric aerosols and clouds. Among the instruments included were a Polly^XT Raman-polarization lidar to retrieve aerosol vertical distribution, a 35-GHz cloud radar to obtain cloud microphysical properties, a disdrometer to measure precipitation, a Doppler wind lidar to track aerosol and cloud dynamics, and a microwave radiometer to measure water vapor and liquid water content.

Furthermore, the TROPOS mobile radiation observatory (MORDOR) complemented the observational capabilities of LACROS by conducting radiation measurements between June 2017 and January 2018. These measurements were integral for quantifying the direct and indirect effects of aerosols and clouds on solar radiation.

2.2

Instrumentation

The instrumentation deployed as part of the MORDOR setup in Limassol was chosen to ensure the high accuracy and reliability of solar irradiance measurements. Central to this setup were two Class A instruments, as classified by ISO 9060:2018 standards, provided by EKO Instruments. These included:

Additionally, to complement these irradiance measurements, a fish-eye camera (Model Vivotek FE8172V) was used to capture sky images every 10 minutes from sunrise to sunset. These images are invaluable for visual assessments of sky conditions, cloud coverage, facilitating a more comprehensive analysis of solar irradiance data.

2.3

Clear sky model

For the analysis of clear sky conditions, simulations of clear sky irradiance were conducted daily throughout the period from June 2017 to January 2018. These simulations employed a 15-minute timestep, with results interpolated to a 1-minute resolution to align closely with the temporal frequency of the measurements. The core of these calculations was the utilization of precalculated look-up tables (LUTs), a strategic choice to enhance computational efficiency.

These LUTs contain simulated solar surface radiation (SSR) values generated using the libRadtran radiative transfer package (RTM)⁸. Simulations were performed for a range of solar zenith angles (SZAs) and atmospheric parameters representative of cloudless conditions, thereby facilitating accurate estimates of clear sky irradiance. Given that this modeling approach, including the specific parameters and setup, has been previously detailed in the literature, here we will not reiterate those specifics. For a comprehensive understanding of the model’s configuration and the parameters employed, readers are directed to the work of Fountoulakis et al.⁹, where these elements are exhaustively described.

2.4

Analysis methods

The first criterion is that the elevation of the sun must be above 10 degrees. The second criterion requires the measured irradiance to be higher than the corresponding clear sky value, and the third insists that the sun-disc should not be obscured by clouds.

To accurately determine the occurrence of CEE, we initiated our investigation by evaluating the performance of our chosen clear sky model. This initial step is very important, as it aids in establishing an appropriate threshold for modeled clear sky irradiances. Exceeding this threshold with measured irradiances is indicative of an enhancement, thereby characterizing CEE. For the identification of clear sky periods within our dataset, we employed the method developed by Inman et al.¹⁰, adapted for MATLAB by Jamie Bright¹¹. This model’s reliance solely on GHI and DNI measurements, alongside their respective clear sky estimates, renders it particularly suitable for our study’s context, where additional data types (e.g., diffuse irradiance) are not available. The model’s applicability in data-limited scenarios, coupled with its demonstrated accuracy in pinpointing clear sky conditions, underscored its selection for our analysis¹². The essential clear sky estimated of GHI and DNI required for the operation of the clear sky detection algorithm were sourced from the Copernicus Atmosphere Monitoring Service (CAMS) solar radiation database ¹³, offering a high temporal resolution of 1-minute intervals.

Following the identification of clear sky periods, our next step was to validate the mode described in section 2.3 against actual irradiance measurements. Given the variability of atmospheric conditions beyond the inputs used in the Radiative Transfer (RT) model, discrepancies between modeled and measured GHI are anticipated. Previous research has typically adopted a fixed threshold over the clean sky (without clouds and aerosols) irradiance—often a 5% difference between measured GHI and modeled clean sky GHI—to delineate CEE (e.g.,^14,15). However, recognizing that our clear sky model incorporates the effect of aerosols, its performance exhibits a dependency on the Solar Zenith Angle (SZA), we opted for an SZA-dependent threshold to refine our CEE identification process.

To establish this tailored threshold, we analyzed the distribution of differences between modeled clear sky GHI and actual measured GHI, focusing on their quantiles. Specifically, we utilized the 90^th percentile of these differences as a preliminary threshold, which acknowledges the slight model performance variation with SZA. To articulate this relationship more precisely and derive a final threshold formula, we fitted a third-degree polynomial to the 90^th percentiles plotted against SZA. The resulting equation for the threshold as a function of SZA is expressed as:

This methodological adjustment for an SZA-dependent threshold is aimed at enhancing the accuracy of CEE detection by aligning the threshold more closely with the empirical performance of the clear sky model across different SZAs.

In addition to the condition where the difference between the measured GHI and the modeled clear sky GHI must exceed the specified threshold, as determined by Equation 1, an additional criterion for a valid CEE occurrence is the presence of an unobstructed sun-disc. This incorporates direct observations of the solar disc’s visibility into our analysis. Consequently, a further essential condition requires that the DNI exceed 120 W/m².

3.1

Detection of clear sky conditions

Clear sky conditions were identified using the method proposed by Inman et al.¹⁰, utilizing 1-minute measurements of GHI and DNI, along with their respective clear sky estimated from the CAMS solar radiation database. Figure 1 illustrates the time series of measured GHI (grey line) during the period from 3^rd to 15^th October 2017, alongside the corresponding clear sky periods detected in the measurements. In this example, the clear sky periods identified demonstrate the selected clear sky detection model’s effective performance.

Figure 1:

Visual representation of clear sky detection during the period 03-15/10/2017. Measured irradiance is in grey, and black thick line signifying minutes identified as clear.

3.2

Identification of enhancement events

To evaluate the performance of our clear sky model, it was compared with the measurement during the detected periods of clear sky conditions, as identified in section 3.1. Figure 2 presents a density scatter plot between the modelled clear sky GHI and the measured GHI for the detected clear sky conditions (a) and the histogram of the absolute differences between the two quantities. Although, they are highly correlated, displaying the model’s ability to follow the patterns of GHI, there are some overestimations of the model. These discrepancies could be related to the inputs of the model, such as the aerosol properties or total precipitable water. Since overestimation is consistent, for the identification of the threshold for the definition of a CEE, this needs to be taken into consideration. Figure 3 shows the differences between the model and the measurements of GHI for clear sky conditions as a function of the SZA. Corresponding boxplot has been made for bins of 5 degrees SZA. To account for the model’s overestimation, we assumed that the threshold for the identification of a CEE is when the difference of measured GHI and the modelled clear sky GHI is above the 90^th quantile (q90). However, since the differences show a dependence from SZA, the 90^th quantiles for each SZA bin was calculated and then a 3^rd degree polynomial as a function of SZA was fitted, which is the final threshold (orange dashed line in Figure 3). Thus in order to have a CEE the measured GHI under all sky conditions minus the modelled clear sky GHI should as a function of SZA should be above the threshold identified by the orange dashed line. Taking as limit the q90 of the clear sky modelled GHI minus the measure GHI, we account for the overestimation in the modelled data, while introducing the SZA we take into account the different behavior for different SZAs. An additional critical factor in this analytical process is the requirement for DNI to be greater than 120 W/m², a criterion grounded in the World Meteorological Organization’s (WMO) current threshold for sunshine duration¹⁶.

Figure 2:

Density scatter plot of the modeled clear sky GHI versus the measurements for the detected clear sky conditions (a), and the distribution of the absolute differences between the modeled and measured GHI for clear sky conditions (b).

Figure 3:

Absolute differences between the modeled and measured GHI for clear sky conditions as a function of SZA. The boxplots show the differences in bins of 5 degrees, the blue dots represent the 90^th percentile (q90) of the differences, and the orange dashed line is a third-degree polynomial fit of the 90^th percentiles as a function of SZA.

The time-series of the measured GHI, along with the respective modeled clear sky irradiance and the CEE as identified above is shown in Figure 4. The data illustrates the maximum recorded GHI of 1109 W/m² during the observation period. CEEs are discernible in 3.6% of total GHI measurements and are present on 49% of the days within our dataset, affirming their non-negligible contribution to solar irradiance. These results in close agreement with the ones reported in a similar study for Thessaloniki, Greece¹⁷, although our dataset is limited to 6 months. However, the observed frequency of CEEs during the summer months is notably lower. This trend can be attributed to the scarcity of cloud cover over Limassol during this period^9,18. The mean excess irradiance for CEEs is approximately 40 W/m², while the maximum observed enhancement reaches 264 W/m². These values suggest that during the autumn and winter months, the irradiance levels during CEEs approach or even surpass those typically observed in the summer months, indicating that CEEs can significantly alter the expected seasonal irradiance profile (Figure 4). Moreover, our analysis reveals instances where the measured GHI exceeds the total solar irradiance at the top of the atmosphere (TOA), adjusted for the horizontal plane, as demonstrated in Figure 6. Out of the CEE occurrences cataloged, there are 24 cases—accounting for approximately 0.5% of all cases—where the GHI during a CEE surpasses the TOA levels.

Figure 4:

Time-series of the measured and clear sky modelled GHI, along with the irradiance enhancement events due to broken cloud conditions.

To further investigate the conditions under which CEE occur, we examined their distribution as a function of SZA and azimuth angle (Figure 5). The variation in CEE frequency with respect to SZA is indicative of the complex interplay between the sun’s position and the atmospheric path length of sunlight. The data suggests an inhomogeneous distribution, with pronounced peaks at low SZAs where CEEs are significantly more frequent. This observation aligns with the hypothesis that certain atmospheric conditions—possibly characterized by cloud types with favorable edge geometry or specific aerosol concentrations—enhance the probability of CEEs through the optimal refraction and reflection of sunlight. However, there is a specific pattern of CEE appearance over specific azimuthal angles, which might suggest that the formation of clouds may happen over specific hours.

Figure 5:

Three-Dimensional distribution of CEE frequencies as a function of solar zenith and azimuth angles.

3.3

Case study

Figure 6 illustrates the diurnal variability of GHI (depicted in blue) for October 7, 2017. Enhancement events, marked by red circles, denote instances where measured irradiance exceeded the threshold. The modeled clear sky GHI is represented by an orange dashed line, illustrating expected irradiance levels under clear sky conditions. The green line indicates the solar irradiance at the top of the atmosphere (TOA), adjusted for the horizontal plane. Notably, during the hours of 8 to 10 AM UTC, several instances were observed where the irradiance during a CEE surpassed the irradiance levels at the TOA, highlighting the significant impact of these events on solar irradiance.

Figure 6:

Diurnal variability of GHI (blue) for 2017-10-07. Red cycles denote the enhancement cases that were identified during the day. The orange dashed line represents the clear sky modelled GHI for this day, and the green line is the solar irradiance at the TOA, adjusted for the horizontal plane.

Presented in Figure 7 is the ACTRIS Cloudnet¹⁹ target classification for October 7, 2017, elucidating the types of clouds present during the occurrence of the day’s CEEs. It is evident that during the peak hours of CEEs (8-10 AM UTC), ice clouds extending from 4 to 6 kilometers were prevalent. The presence of these ice clouds is implicated in the enhanced reflection of solar irradiance, contributing to the observed over-irradiance events. This figure underscores the critical role of specific cloud formations in the amplification of solar irradiance at the surface, offering insights into the atmospheric conditions conducive to CEEs.

Figure 7:

Cloudnet target classification for 2017-10-07.