Long-term series of surface solar radiation at Athens, Greece

17 We present a long-term series of solar surface radiation (SSR) for the city of Athens, Greece. The 18 SSR measurements were performed from 1953 to 2012, and before that (1900-1952) sunshine 19 duration (SD) records have been used in order to reconstruct monthly SSR. Analysis from the whole 20 dataset (1900-2012) mainly showed: a decrease of 2.9% per decade in SSR from 1910 to 1940 21 assuming a linear change in SSR. For the dimming period (1955-1980), a -2% change per decade 22 has been observed, that matches various European long-term SSR measurement related studies. 23 This percentage for Athens is in the lower limit, compared to other studies for the Mediterranean 24 area. For the brightening period (1980-2012) we have calculated a +1.5% per decade which is also 25 in the lower limit of the reported positive changes in SSR around Europe. Comparing the 30-year 26 periods (1954-1983 and 1983-2012) we have found a difference of 4.5%. The difference was 27 observed for all seasons except winter. Using an analysis of SSR calculations of all sky and clear 28 sky (cloudless) conditions/days, we report that most of the observed changes in SSR after 1954 can 29 be attributed partly to cloudiness and mostly to aerosol load changes. 30 31 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2017-48, 2017 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 14 March 2017 c © Author(s) 2017. CC-BY 3.0 License.


Introduction
In the past decades surface solar radiation (SSR) and the transmission of the atmosphere have been of increasing interest because of the related impacts on climate.Most of the energy in the Earthatmosphere system is introduced by solar radiation as it provides heating, which creates pressure gradients and ultimately wind, as well as it triggers water, carbon and oxygen cycles through evaporation and photosynthesis.These processes define the climatological conditions, and changes of incoming solar radiation rapidly affect the energy balance (Wild et al., 2015).Interest on the solar radiation changes has also been raised after the development of solar energy applications, which are continuously growing in number over the recent years.Changes in SSR have been recorded over the last century and can be caused either by natural events such as volcanic eruptions or human-related activities, mainly in polluted regions (Wild, 2016).At larger scales (thousands of years) changes in SSR, might have been caused by changes in the Earth's orbit and Sun solar output (Lean, 1997;Ohmura, 2006).
Systematic continuous measurements of SSR were established in the middle of the 20th century at selected meteorological observatories.Solar variations have been investigated in several studies using ground based SSR measurements from various monitoring networks worldwide (e.g., Ohmura, 2009) and also by satellite-derived estimations (e.g.Kambezidis et al., 2010).Overall, most of these studies (Gilgen et al., 1998;Noris and Wild ,2009;Wild, 2009 and2016 andreferences therein) have reported a worldwide decrease of solar incoming radiation in the period 1960-1985 (known as dimming period), followed by an increase (brightening period) thereafter.
These changes were larger in more polluted and urban areas but have also been recorded in isolated regions such as the Arctic (Stanhill, 1995) and Antarctica (Stanhill and Cohen 1997).Changes in atmospheric transmission due to variations in cloudiness and aerosol concentration are the main factors to be investigated in order to determine the possible causes of such trends in SSR (Wild, 2009).
The cloud and aerosol radiative effects on solar radiation variations over the past decades have been investigated by numerous studies during the last years.The inter-annual variations in cloudiness is crucial for studying SSR time series, but its decadal variability is not always connected with the widespread dimming and brightening effects (Wang et al., 2012;Wild, 2016).Aerosols play significant role in incoming radiation, by scattering and absorbing light and by acting as cloudcondensation nuclei.Over the 20-year dimming phase (from 1960 to 1980) and the 15-year brightening phase (from 1990 to 2005), it was found that the aerosol effects (direct and indirect) played the most important role in SSR variation (Dudok de Wit et al., 2015).Concerning Central Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2017-48, 2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.
Europe, Ruckstuhl et al. (2008) suggested that the brightening phase under cloud-free conditions is in line with decreasing anthropogenic aerosol emissions (Streets et al., 2006).Nabat et al., 2013 using a blending of remote sensing and model products showed that a decreasing Aerosol Optical Depth (AOD) trend of 0.05 per decade has been calculated for Europe.In addition, Nabat et al., 2014 reported that anthropogenic aerosol decline in Europe from 1980 to 2012 statistically explains explain 81 ± 16% of the observed brightening.Overall, changes in anthropogenic aerosol emissions are now considered as the major cause of brightening and dimming effects (Wild, 2016).The gaseous and particulate air pollutants may reduce solar radiation by up to 40% during air pollution episodes (Jauregui and Luyando, 1999).This attenuation is much larger during forest fires, dust events and volcanic eruptions.Vautard et al., (2009), have also reported on a decline of the frequency of low-visibility conditions such as fog, mist and haze in Europe over the past 30 years, suggesting a significant contribution of air-quality improvements Long-term series of SSR measurements are essential for such studies.One of the main constraints in studying SSR temporal changes is the small number of sites with reliable long-term records, even over areas with high density of stations such as Europe, Japan or the USA.In Europe for example, there are currently less than 80 stations with more than 40-years homogeneous data (Sanchez-Lorenzo et al., 2015), with very few of them operating over Southern Europe.Recently, a highquality dataset of SSR has been set up over Italy (Manara et al., 2016), but there is still lack of high quality long-term trends in other countries around the Mediterranean Basin.
In addition, even more sporadic measurements are available before the 1950s (Stanhill and Achiman, 2016); the few studies of them have pointed out an SSR increase in the first decades of the 20 th century and a maximum around 1950 (Ohmura, 2006).This topic is still controversial due to the few long-term series available (Antón et al., 2014).Recently, there have been efforts to reconstruct SSR series in periods with no direct measurements available, using other variables such as sunshine duration (SD), which is available in a large number of sites since the late 19 th century (e.g., Stanhill and Cohen, 2005, for USA;Sanchez-Lorenzo and Wild 2012, for Switzerland;Matuszko 2014, for Poland).For example, Sanchez-Lorenzo and Wild (2012) used data from 17 stations in Switzerland, considered SD as a proxy and successfully reconstructed SSR time series since the late 19 th century.Thus, they calculated that the variability in SSR monthly anomalies can be explained by SD anomalies in a range of 76%-96%, and a monthly root mean squared error of 4.2 W m -2 between recorded and estimated SSR for all-sky conditions and of 5.5 W m -2 for clearsky conditions.Other studies have tried to use pan evaporation as a proxy of SSR, for the first half of the 20 th century (Stanhill and Möller, 2008).Kambezidis et al. (2016)  For the Southeastern Mediterranean, there have been a few studies discussing the brightening/dimming effect.Zerefos et al. (2009) have studied the Ultraviolet A (UVA) changes for the area of Thessaloniki (Greece) from 1984 to 2008.They calculated a 5% positive trend per decade linked to a negative trend in aerosol optical depth (AOD) for the area due to air pollution abatement strategies.Founda et al., (2014) have studied the SD long-term variability over Athens area.They reported a 7% decline in the annual SD from 1951-1982 and a 3% increase from 1983-2011 under all sky conditions.Although, under near clear sky conditions, these percentages are -7% and + 9% for the dimming and brightening periods respectively.Similarly, Founda et al. (2016a) analyzed long-term SD and total cloud cover time series over 15 sites in Greece (the oldest one beginning on 1897).They have shown an increase in SD almost at all stations since the mid-1980s, which in certain areas of Southeastern Greece amounts to an increase of 20 h per year.This increase is not accompanied with synchronous decrease in total cloud cover, possibly evidencing to decreasing aerosols loads, despite the fact that their impact on SD should be lower than on SSR (Sanchez-Romero et al., 2014).Yildirim et al. (2014) have analyzed 41 years of SD measurements in 36 stations in Turkey.They reported a decreasing trend (between 1970 to about 1990) for most of the stations.After 1990 they observed either zero trend variation or a reduction in the decreasing rate of SD for most of the locations.They concluded that the decreasing period might be attributed to human-induced air pollution.Founda et al. (2016b)  In this work, measurements of SSR, recorded for 60 years at the center of Athens, are presented.In addition, with the use of the SD measurements that are conducted in Athens since 1900, we could reconstruct the time series of SSR during the first half of the 20th century.These time series  are the oldest, uninterrupted and high quality SSR time series in the SE Mediterranean and one of the oldest in Europe, providing unique information about the variations and trends in the area for the past decades.Time-series of SSR over Athens are presented to try answering questions such as: Atmos.Chem.Phys. Discuss., doi:10.5194/acp-2017-48, 2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.
Are the dimming-brightening patterns observed in Europe over the past century also observed, at the same extent, over the Eastern Mediterranean?
Is SSR variability during the first decades of the 20th century in Athens in line with the other few locations reporting trends over this period?
Can we verify that anthropogenic aerosols play the most important role on the brightening/dimming observed SSR after 1950 in agreement with other European regions?
2 Data and Methodology

Data collection and analysis
The SSR data used in this study cover the period from December 1953 to December 2012 and were measured by a series of pyranometers that are mentioned in Table 1.These instruments have been operating continuously at the Actinometric Station of the National Observatory of Athens (ASNOA) (Hill of Pnyx, Thissio), that is located near the center of Athens, Greece (38.00 o N, 23.73 o E, 110 m above mean sea level).Table 1 presents the instruments and the period of operation, as well as the maximum error in the calculation of the daily values.References mentioned in Table 1 describe the exact type of errors and uncertainties related to the sensors.In the period 1953-1986, the maximum daily error was about 5%, and 2% afterwards.The spectral response of the sensors is in the range of 285-2800 nm; since 1986 a first-class Eppley PSP pyranometer (WMO, 1983) is operating at ASNOA.Since 1992, frequent calibrations (every two years) have been performed by the NOA's Laboratory of Meteorological Device Calibration (LMDC, 2016) in order to ensure the high quality of measurements.LMDC follows the standard calibration procedure for thermopile pyranometers (ISO 9847, 1992), with exposure to real sunlight conditions and comparison with a standard thermopile (Secondary Standard) pyranometer.LMDC's reference pyranometer, Kipp & Zonen CM21, is regularly calibrated in PMOD/WRC, Davos, Switzerland. SSR values greater than 120% of the seasonally corrected solar constant.
After the initiation of diffuse horizontal radiation measurements at ASNOA in 1991, the following quality criteria were added for rejection:  diffuse horizontal values greater than the corresponding SSR ones;  diffuse horizontal values greater than 80% of the seasonally correct solar constant;  direct-beam solar component exceeding the extraterrestrial solar irradiance.In addition, collocated measurements of SD recorded at ASNOA have been used.According to WMO (2010), the SD during a given period is defined as the sum of the sub-periods for which the direct solar irradiance exceeds 120 W m -2 .In Athens, SD has been recorded using classical Campbell-Stokes heliographs (since 1894) and been replaced by electronic instrumentation in 1998 (EKO, MS-091 analog SD sensor).Monthly SD values since January 1900 have been used in this study.A more analytical study of these time series can be found in Founda et al. (2014).
Complementary to this study, cloud-cover observations from the Hellenic National Meteorological Service (HNMS) from 1954 have also been used.These observations are recorded at a site 7 km away from ASNOA.All cloud observations at HNMS are conducted every 3 hours and are expressed in octas.
Concerning the data availability for SSR and SD data: SSR monthly means calculated here have been retrieved from daily calculated SSRs.Over the 59 years (708 months), 98% of the months had none or one day missing, 3 months had from 10-20 missing days and 2 months from 20-30 missing days.For SD, 1931-1940 monthly data have been used taken from the NOA measurement annals.
From 1940 on, hourly measurements have been used in order to derive daily and monthly measurements.The SD time series have no gaps with only six missing days during December 1944 (Founda et al., 2014).
In order to examine the AOD impact on SSR, we have used the longest satellite based AOD series available for the area.This is the AOD time series from Advanced Very High Resolution Radiometer (AVHHR).AOD retrievals at 630 nm over global oceans at spatial resolution of 0.  (Mishchenko et al, 2007).Current dataset radiances have been recalibrated using more accurate MODIS data (Chan et al, 2013).We used daily data at the region around Athens

Clear-sky SSR
For the determination of the clear sky (defined here as the cloudless) days, we have used both the cloud octas and SD data.Daily observations have been used for this analysis.We have defined as a clear sky day each day that fulfills the following criteria: -the mean daily cloudiness (in octas) should be less than 1.5, and -the total daily SD should be higher than 90% of its theoretical (astronomical) value.
The procedure for calculating a single mean cloud octa value for each day was the following: We have first excluded night-time cloud observations; then, we have weighted each observation based on the hour of the observation.Weights have been calculated based on the solar radiation contribution of the specific time slot and day of the month, compared with the daily clear sky SSR integral, of the particular day and month.

Reconstruction of SSR from SD
We have used the 1900-2012 SD time series in order to extend our SSR time series back to 1900.
For that purpose we have used a recent period  in order to derive a function between SD and SSR and a testing period  to verify the validity of this function and results.
Finally we applied this derived and tested formula for the 1900-1952 period where SD measurements were available while SSR were not.In order to derive a relationship between SD and SSR, we used the broadly accepted formula of Ångström (1924): where SSR max and SD max refer to the theoretical extra-terrestrial value of SSR and the astronomical value of SD, respectively, while a and b are constants usually defined monthly.This formula can 3 Results

Relationship between SD and SSR measurements
In order to benefit from the over-century long time-series of SD at NOA we used Eq. ( 1) for the period of synchronous SD and SSR measurements to estimate constants a and b for the period 1983-2012.The 1983-2012 period was chosen for determining the SSR vs. SD relationship as mainly SSR measurements have lower uncertainties compared with the 1953-2012 one.We thus calculated a=0.237 and b=0.458 (as derived from the linear equation shown in Figure 2) and used them for validation in the period 1953-1982.Figure 2 shows the correlation of the monthly SSR/SSR max and SD/SD max ratios for the 1983-2012 period, together with the derived coefficients (a and b) and the coefficient of determination (R 2 ).In addition, we have included the testing period statistics, together with the coefficients a and b that can be determined using the whole period , to show that each of the two periods could provide similar results.SSR values.The latest can be directly linked with the Pinatubo volcanic eruption and its known effect in the SSR (e.g.Zerefos et al., 2012).Analytical linear trends of each of the sub-periods and for every season are presented in Table 2. Looking at the 1900-2012, period the seasonal and annual trends in SSR are less than 1% per decade.A positive change of 0.40% per decade has been calculated from annual values.For the whole data set, all seasons show positive trends, except for winter.For the periods with simulated SSR values , even smaller trends have been detected for spring and summer.The measuring period of 1954-2012 has been split into two sub-periods of 1954-1982 and 1983-2012.
The first sub-period shows a negative annual change of -2.33% per decade in SSR, which is also reflected in all seasons with predominant changes during winter (-6.43% per decade).The second sub-period shows a positive trend of +0.80% per decade with highest ones in spring (+2.77% per decade) and summer (+1.38% per decade) and negative in autumn (-1.50% per decade).Looking at the trend significance described by the 95% confidence bounds, we can see significant positive trends for 1900-2012 (yearly, summer and spring) and significant negative trends for yearly analysis and winter of 1953-1982.
In order to have a better understanding of the SSR changes over the 112-year period (1900-2012), we have calculated the decadal SSR trends for different time-windows (15 to 40 years).Figure 5 shows the results of this analysis.For the first two decades of the 20 th century there appears a decrease in SSR, in line with other long-term SD series as recently shown by Stanhill and Achiman (2016).Then, in all calculations an increase is shown from mid 1930's to late 1940's, in line with the early brightening effect pointed out by other authors (Ohmura, 2009;Sanchez-Lorenzo et al., 2008).It should be reminded that this period is based on estimations of SSR from SD measurements, which thus include additional measurement uncertainties.Nevertheless, both the early dimming and brightening periods reported in Stanhill and Achiman (2016) and this study seem to be in line with trends in anthropogenic black carbon (McConnell et al., 2007;Lamarque et al., 2010) and biomass-burning (Lamarque et al., 2010) emissions peaking in the 1920's and then decreasing.The dimming period from 1950's to 1970's can be observed in all time windows with a brightening effect after late 1970s.
The 40-year and 30-year time windows in the analysis presented in Figure 5 show the maximum rate of increase in early 1940's (resulting in an increase of 2% per decade and 3% per decade, respectively).Then a maximum rate of decrease is observed in early-mid 1960's, followed by a positive rate of increase after 1990's.Shorter time windows (15 years) are also interesting as they are able to capture the Pinatubo effect in early 1990's.-periods: 1954-1983, and 1983-2012 (the common year is meant to have equal duration for both periods).Investigating a possible seasonal dependence, the relative difference in SRR for every month from its mean monthly value of the whole measurement (1953-2012) period was calculated.
Figure 6, shows the mean daily insolation for each month for the two sub-periods and the whole 59year period.Examining the monthly average differences between the two periods, we observe that for spring and summer months these are of the order of 6%.In addition, for all months SSR differences of the 1983-2012 period compared to the 1954-1983 period are positive with an exception of November (-1.9%) and December (-1.2%).In general, the second measurement period shows a 3% to 8% larger monthly SSR than the first measurement period.We have also calculated decadal trends in time windows of 15 to 30 years for the entire SSR measurement period (see Figure 7).In the left panel of Figure 7

Comparison between all-sky and clear-sky SSR records variation
We have used the 59-year data set  in order to quantify the factors controlling the SSR variations in Athens, Greece, focusing mainly on two known dominant factors, clouds and aerosol load.Figure 8 suggests anti-correlation between variations in SSR and cloud cover.Figure 9 shows the correlation between annual mean SSR and cloud cover.From the best-fit linear regression line it is seen that a -1.54 MJ m -2 (or -9.6%) change in mean daily insolation accounts for a change of 1 octa in cloud cover.In addition, there is no significant change in cloudiness over the 59 year period for Athens, Greece.

The role of clouds
Calculating linear changes of cloudiness from data shown Figure 8, shows a non significant change of -0.4% per decade which can practically have a limited effect on SSR changes during the examined period.
Nevertheless, it is worth mentioning that different cloud properties like cloud optical thickness and cloud phase, not described by the measurements of cloud cover, can influence SSR.

Clear sky records
In order to minimize the cloud influence and investigate the possible role of direct aerosol effects on Athens SSR series, we had to select clear-sky (or cloudless) days.We have used daily SSR measurements from 1953 to 2012 and we have separated the cloudless days according to the criteria mentioned in Section 2.2.
For considering the SSR seasonality, we have calculated a five-degree polynomial derived from the maximum daily SSR (for all years of the data set), as a function of the day of the year (Figure 10).
Afterwards we have calculated the ratio of the daily SSR to the SSR calculated by this function.
Seasonal and yearly means of this ratio have been estimated and have been used to describe cloudless-sky SSR percentage changes on a seasonal and yearly basis.This approach has been Figure 11 shows that most of the SSR variation observed for the measuring period has to be explained by other factors than changes in cloudiness (see figure 8 for variations due to cloudiness).
Different seasons with the exception of wintertime show similar patterns to the year-to-year variability.Individual seasonal calculated SSRs do not exceed by more than ±5%, the SSR variability of all sky data, with the exception of the winter season.Comparing clear sky and all sky yearly mean SSR, we find a high correlation (R 2 = 0.71), which can be explained as a combination of: aerosol changes driving the SSR changes and by the number of clear sky days during the year.
There is a decrease of more than 15% in the clear sky SSR from the start of the series to the end of 1960's.A decline after 1983 could possibly be related with El Chichon volcanic eruption.
The Pinatubo-related drop of 10% from the early 1990's to the mid 1990's, can also be seen in both cloudless and all-sky datasets and also to the increase in AOD in the AVHRR dataset (Figure 11).
Finally, the ~13% change from 1995 to 2012 shown for all skies (Fig. 8) and clear skies (Fig. 11) is accompanied with a drop of ~25% in AOD measured by AVHRR.The year to year variations of clear sky SSR series and the AVHRR-related AOD show an anti-correlation with R=-0.78 (N=29), verifying the hypothesis that SSR clear sky changes are associated with aerosol load changes, at least within the common AVHRR/measurement period .Differences between the AVHRR and ChaArMEx data can be explained in part by the different AOD wavelengths presented here (630 vs 550 nm) and also by a general negative bias of AVHRR over the Mediterranean compared to AERONET (Nabat et al., 2014).The smooth decline in the ChArMEx AOD data is due to the method used to build this product and uses the trend and not the interannual variability which is not included in the global model that was used.Figure 12.Ratio of cloudless vs all days, per season and yearly In addition to figure 11 we have included figure 12 showing the ratio of cloudless days to all available days for each season and for each year.Figure 12 shows a minimum (less than 30% during a year) of the number cloudless days from mid-1970's to early 1990's.It is mostly linked with the decrease of cloudless days during summer months.The figure provides a hint on the SSR relative changes observed during this period, but it can not directly interpret year to year SSR changes as they depend also on cloud fraction and properties for cloudy days.In addition, it can only partly be linked with fig.11 as aerosol effects on cloudless sky calculated SSRs depend mostly on AOD levels and not on the number of days included in the calculations.
In Table 3 we have calculated the linear trends for the 1953-2012 period and for both clear sky and all sky measurements and the 1953-1982 and 1983-2012 sub-periods for clear sky measurements.
Results show comparable changes per decade (2% for the clear sky and 1.5% for the all sky cases).
Seasonal analysis show that clear sky trends for summer, autumn and winter months are higher than the ones derived for all skies.Such differences are linked with the seasonal variability and longterm changes of cloudiness for the specific seasons.De-seasonalized SSR data analysis from 1900 to 2012 showed high month to month variability that could reach up to 25%, mainly related with monthly cloudiness variations.During the period 1910mid-1930s where only few datasets have reported worldwide SSR results, we observe a 2.9% per decade or a total of 8.7 % decrease in SSR, assuming linear changes in SSR during this period.This early dimming was followed by a 5% per decade increase from 1930 to the 1950s.Similar results have been found at Washington DC and at Potsdam, Germany (Stanhill and Achiman, 2016) Stanhill and Achiman, 2014This work 1959-19881971-19861959-19851953-1982 Europe Europe Italy Athens, Greece -2.0 -2.3 -6.4(±1.1)/ -4.4(±0.8)-2.33(±2.28)Ohmura and Lang, 1989Norris and Wild, 2007Manara et al, 2016This work 1985-20051990-20121986-2013 Europe Italy Athens, Greece 2.5 6.0 (±1.1) / 7.7 (±1.1) 0.80 (±1.96)Wild, 2009Manara et al, 2016 This work The decadal variations of SSR measured since 1954 at Athens, Greece, originate from the alterations in the atmosphere's transparency (namely by clouds and aerosols).Using an analysis of SSR calculations of all sky and clear sky (cloudless) days we end up that since cloud cover changes during the 59 period were very small, most of the observed decadal changes might be related with changes in the aerosol load of the area.An additional hint in support of this conclusion is the high correlation of clear sky and all sky yearly SSR.We also found an anti-correlation between either clear sky and all sky SSR measurements and AOD time series from AVHRR  or ChArMex (1979ChArMex ( -2012)).Looking at linear trends over the 59 year period clear sky changes per decade were 2% while it was 1.5% for all sky conditions.The most pronounced changes have been calculated for summer and autumn seasons (2% and 2.7% respectively).
used monthly re-analysis Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.datasets from the Modern Era Retrospective-Analysis for Research and Applications (MERRA) and calculated shortwave radiation trends from 1979-2012 for the Mediterranean basin.They reported an increase in MERRA by +0.36 W m −2 per decade, with higher rates over the western Mediterranean (+0.82 W m −2 per decade).
have investigated the visibility trends over Athens area from 1931 to 2013.They reported a deterioration in the visibility up to 2004 and a slight recovery afterwards, negatively/positively correlated with relative humidity/wind speed and positively correlated with AOD from 2000 to 2013.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.Also, both total and diffuse horizontal measurements are corrected for the night-time dark-signal offset of the pyranometers.Mean daily SSR values were calculated from the data set of this study (December 1953 -December 2012); only months with more than 20 days of measurements were considered in the analysis.Over the 60 years of measurements, only three months (January and February of 1998 and March of 2012) did not fulfill this criterion.

Figure 1
Figure 1 shows the intra-annual variability of SSR at ASNOA based on the measurements from all instruments during the period 1953-2012.Mean daily SSR at Athens ranges between approximately 6 to 27 MJ m -2 during the year.Mean and standard deviations were calculated using the 60 year record for each day.

Figure 1 .
Figure 1.Average Intra-annual variability of Surface Solar Radiation (SSR) at Actinometric Station of the National Observatory of Athens (red), along with the inter-annual variability for a given day (±1 standard deviation, blue), calculated over the period 1953-2012.

(
longitude: 37.5º-38.2ºN,latitude: 23.2º-24.4ºE)which includes 50 active available (ocean) gridpoints.The above region was selected based on data availability on each grid with the distance up to 50 km from ASNOA.Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.To complement the analysis on the evolution of aerosols, the recent climatology developed by Nabat et al., 2013 has been considered over the period 1979-2012.This product provides monthly averages of AOD at 550 nm over the Mediterranean region at 50 km resolution.It is based on a combination of satellite-derived (MODIS instrument) and model-simulated products (MACC reanalysis and RegCM-4 simulations), which have been selected among many available datasets, from an evaluation against ground-based measurements of the AERONET network.Thus this climatology is able to give an estimation as best as possible of the atmospheric aerosol content over the period 1979-2012.For the present work, the AOD time series over the grid cell of the ASNOA (38.00 o N, 23.73 o E) has been extracted and is referred to as the ChArMEx data thereafter.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017   Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.only be used in large data sets as a statistical approach.That is because for different cloud height, thickness and positioning, the constants can show a large variability(Angell, 1990).

Figure 2 .
Figure 2. Plot of Y=SSR/SSR max vs X=SD/SD max .Blue dots represent the 1953-1982 period and red ones 1983-2012.The red and blue lines represent the respective linear regression lines.Corresponidng regression relations are given in the inner boxes.For testing the method we have applied the retrieved coefficients and calculated the SSR for the 1953-1982 period.Comparing normalized simulated with the observed SSR values an R 2 =0.79 has been found (see Figure3).

Figure 4 .
Figure3.Plot of the ratios of mean monthly SSR values to SSR maximum, of measurements in the period 1983-2012 vs the estimated ratios from Eq (1) for the same period.Blue is the 1:1 and black is the regression line.

Figure 5 .
Figure 5. Trends in SSR (% per decade) calculated for different sliding time windows.The value of the trend has been calculated at the central year of each time window.
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017   Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.3.3Variations and trends in SSR for the 1953-2012 measurement periodIn order to further analyze the whole 59-yr SSR data set of this study, we have divided it in two 30yr climatological sub

Figure 6 .
Figure 6.Intra-annual variability of monthly mean daily SSR over the sub-periods of 1954-1983 (blue line) and 1983-2012 (red line) and the entire period of 1954-2012 (black line).The green line (right axis) represents the monthly relative difference between the two 30-year sub-periods.
each of the points used for constructing the colored lines represents the percent change per decade of the SSR.Mostly positive trends are detected using any time window centered after 1975.Larger trends are calculated for time Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.windows centered at 1975 to 1980 and after 2000 (in the order of 5% per decade using the 15-year time window).For the period 1954 to 1970 mainly negative trends are shown.

Figure 7 .
Figure 7. Left panel: calculation of SSR changes in % per decade for different time scales (15 to 30 years) used.Each point on the line represents the middle year of the time window used for the trend calculation Right panel: % change per decade for different time scales and different months using 1954 as the starting year (the last 30 year period is magnified with changes presented as % per decade).

Figure 8
Figure 8 shows the 1954-2012 time series of the monthly and yearly anomalies based on daily SSR, together with yearly total cloud coverage in weighted octas.The yearly de-seasonalised SSR values for all-sky conditions show a drop of ~14% from 1960 to 1970 and then a continuous increase

Figure 8 .
Figure 8. De-seasonalised monthly mean (i.e.deviation from the respective monthly mean calculated over the whole period; dash line) and yearly mean SSR (black line), along with mean annual total cloud cover (green line).

%
Figure 9. Correlation between annual means of daily insolation and cloud cover over the period 1954-2012.The straight line represents the best-fit regression line to the data points.The year 1953 has not been included in the analysis since it does not include measurements for all months.However, the great scatter of the data points and the low correlation of the two parameters in Figure 9 (R 2 =0.229) indicate that the presence of cloud cover can only partly explain the changes in SSR.
Figure10.Clear-sky SSR measurements (red dots) and all-sky SSR measurements (blue dots) derived with the cloud octa (cloudiness<1.5) and sunshine duration (SD>0.9)related criteria.The black line represents the polynomial fit to the daily SSR max values.
Similar to the AVHRR data the ChArMEx 4-D aerosol climatology is shown in figure 11, providing similar conclusions, namely the AOD negative trend of 0.03 or 14% per decade from 1979 to 2012.% Change of clear sky SSR Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.
Clear sky results for the 1953-2012 period show significant positive changes in SSR for all seasons except winter.Looking individually at the1953-1982 and 1983-2012  periods we have calculated significant negative trends only for the winter over the first and for summer and spring over the second.The effect of various parameters on SSR has been discussed byKambezidis et al. (2016) in their study about the global dimming/brightening effect over the Mediterranean in the period 1979-2012; they show that the influence of parameters related to the atmospheric transparency, like water vapor, aerosols and trace gases, as well as changes in the surface albedo on SSR have been larger in the southern parts of the Mediterranean, over the Balkan countries and central Turkey.This outcome is in agreement with the conclusion of the present study that other factors than cloudiness play significant role in the SSR variations.5 ConclusionsSurface solar radiation (SSR) at National Observatory of Athens, in the center of the city, is presented using a unique dataset covering a period of 59 years.Sunshine duration (SD) records for another 54 years have been used as a proxy to reconstruct SSR time series for the period from 1900 to 2012.A comparison of the SSR results in Athens with visibility observations since 1931(Founda et al.,    2016b)  did not show any correlation among SSR and horizontal visibility.The steep visibility Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.decrease from 1931 till early 90's is not accompanied by a relative SSR decrease excluding individual sub-periods.Studying the literature for similar cases, similar conclusions have been drown byLiepert and Kukla (1997) showing an SSR decrease over 30 years of measurements accompanied by a visibility increase and no significant changes in the cloud cover conditions, in Germany.This Athens SSR vs visibility relationship can be partly explained by the fact that: SSR and visibility have different response on cloud conditions, water vapor and rainfall and also by the fact that visibility is affected by aerosols only in the first few hundred meters above the surface, while SSR is affected by the columnar AOD, which in the case of Athens can be significantly different due to aerosol long-range transport in altitude (e.g.Saharan dust;Léon et al., 1999).The data accuracy of such historic radiation measurements is generally not well established, at least compared with the measurements after the 1990s.Quality assessment procedures in the presented time series have been applied based on criteria based on instrument characteristics and the availability of additional collocated measurements.Year to year fluctuations of the measured SSR in addition to the reversal of the downward tendencies at the Athens site adds credibility to the measured variations.That is because a typical radiometer behavior is to lose sensitivity with time indicating spurious downward, but not upward trends.The more recent (after 1989) SSR measurements can be characterized as high-quality radiation data with known accuracy.

Table 1 :
History of SSR instruments used at ASNOA.SSR measurements refer to the total solar radiation on a horizontal surface.

Table 2 .
Annual and seasonal SSR trends in percent per decade over the period 1900-2012 and different sub-periods.Percentages in parenthesis show the limits of the 95% confidence bounds.

Table 3 :
Clear sky and all sky data trends comparison for the whole 1953-2012 period and the two 30-yr sub-periods (% per decade).Percentages in parenthesis show the limits of the 95% confidence bounds.
. They have reported an early brightening at both locations in the 1930's.For the SSR measurement period of 1953 to 1980, European related studies presented in Wild (2009) showed a -1% down to -7% change per decade in SSR measurements over various European sites (dimming period).For the Mediterranean region, Manara et al. (2016) showed a decrease in the order of -2% to -4% per decade in Italy.We are reporting a change in SSR of -2% per decade in Athens.Finally, for the brightening (1990-2012) phase again Wild et al. (2009) reported a 1.6% up to 4.7% per decade positive change in SSR while we have calculated a +1.5% per decade, which is lower than the 3-6% per decade reported in Manara et al. (2016) for Italy.A summary of the above findings can be seen in table 4. Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-48,2017 Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 14 March 2017 c Author(s) 2017.CC-BY 3.0 License.