Atmospheric Chemistry and Physics

The Earth's climate is driven by surface incident solar radiation (Rs). Direct measurements have shown that Rs has undergone significant decadal variations. However, a large fraction of the global land surface is not covered by these observations. Satellite-derived Rs has a good global coverage but is of low accuracy in its depiction of decadal variability. This paper shows that daily to decadal variations of Rs, from both aerosols and cloud properties, can be ac- curately estimated using globally available measurements of Sunshine Duration (SunDu). In particular, SunDu shows that since the late 1980's Rs has brightened over Europe due to decreases in aerosols but dimmed over China due to their increases. We found that variation of cloud cover determines Rs at a monthly scale but that aerosols determine the variabil- ity of Rs at a decadal time scale, in particular, over Europe and China. Because of its global availability and long-term history, SunDu can provide an accurate and continuous proxy record of Rs, filling in values for the blank areas that are not covered by direct measurements. Compared to its direct measurement, Rs from SunDu appears to be less sensitive to instrument replacement and calibration, and shows that the widely reported sharp increase in Rs during the early 1990s in China was a result of instrument replacement. By merg- ing direct measurements collected by Global Energy Bud- get Archive with those derived from SunDu, we obtained a good coverage of Rs over the Northern Hemisphere. From this data, the average increase of Rs from 1982 to 2008 is estimated to be 0.87 W m 2 per decade.


Introduction
Solar radiation drives the Earth's climate system.The amount that is incident at the surface, denoted R s and measured by a surface network of radiometers, has been shown to undergo significant decadal variations (Gilgen et al., 1998;Liepert, 2002;Long et al., 2009;Ohmura, 2009;Stanhill and Cohen, 2001;Wild et al., 2005).In particular, direct measurements with limited coverage show that the dimming trend of R s that occurred up to the late 1980's reversed to a widespread brightening after that (Wild, 2009;Wild et al., 2005).Satellite-derived R s (Pinker et al., 2005) has a better spatial coverage but may have spurious variability resulting from changes of satellites, their sensor calibration, and undetected low clouds (Evan et al., 2007).It also can be biased by its exclusion of variations of tropospheric aerosols over land (Wang et al., 2009).These changes in atmospheric aerosols contributed to the observed decadal variation in R s over Europe (Norris and Wild, 2007) and Asia (Norris and Wild, 2009).It is not surprising that the decadal variations in satellite-derived R s have been substantially different from those of radiometer measurements at the surface (Hayasaka et al., 2006;Xia et al., 2006).
This paper improves the surface-based global coverage of estimates of R s by using Sunshine Duration (SunDu), a measurement initiated 150 yr ago (Sanchez-Lorenzo and Wild, 2012), which records the time during a day that the direct solar irradiance exceeds 120 W m −2 .It is one of the oldest and most robust measurements related to radiation (Wild, 2009).Worldwide direct R s measurements only began in the late 1950's.SunDu observations provide a globally-distributed proxy record of R s at both urban and rural sites with a much higher density than that of the direct measurements as obtained from the of Global Energy Budget Archive (GEBA) (Gilgen et al., 1998).Consequently, these observations allow us to fill in the blank areas of the GEBA coverage.
Several previous studies have established good correlations between SunDu and R s (Essa and Etman, 2004;Hoyt, 1977;Sanchez-Lorenzo et al., 2009;Stanhill and Cohen, 2005).SunDu has been regarded as a measure of cloud cover.However, studies have shown that aerosols can also reduce SunDu (Horseman et al., 2008;Kaiser and Qian, 2002;Raschke et al., 2006;Sanchez-Lorenzo et al., 2008, 2009;Tang et al., 2011).This paper further demonstrates that SunDu can be used to estimate R s globally with sufficient accuracy to establish its year-to-year variations.It then examines the year to decadal variations of R s and their potential causes.

Data
SunDu records the time during a day that direct solar beam irradiance exceeds 120 W m −2 .It was initiated 150 yr ago and is one of the oldest and most robust measurements related to radiation (Wild, 2009).In 1962 the Campbell-Stokes sunshine recorder was recommended as the reference SunDu sensor by the World Meteorological Organization (WMO) in order to homogenize the data of the worldwide network for SunDu (WMO, 2008).
One advantage of SunDu is that the impact of sensor replacement on SunDu measurement is rather small.Three different types of sunshine duration recorders were used from 1888 to 1987 in the USA (Stanhill and Cohen, 2005): first, the Jordan photographic recorder from 1888 to 1907; second, the Maring-Marvin thermometric sunshine recorder from 1893 to the mid-1960s; and third, the Foster photoelectric Sun Switch (beginning in 1953).Replacement of the recording method has been shown to have a negligible effect on the annual SunDu (Stanhill and Cohen, 2005).
The World Meteorological Organization (WMO, 2008) requires that hours of sunshine should be measured with an uncertainty of ± 0.1 h and a resolution of 0.1 h.There has been no standardized method to calibrate SunDu detectors.For outdoor calibration, the pyrheliometric method was recommended as a reference method, which detects the transition of direct solar irradiance through the 120 W m −2 threshold (WMO, 2008).
SunDu has been regarded as a measure of cloud cover as direct radiation is generally lower than 80 W m −2 for scattered clouds (cumulus, stratocumulus) (WMO, 2008).High and thin cirrus, as well as aerosols only reduce SunDu at low solar elevations, i.e., at times when the incident clear sky solar radiation is not much larger than 120 W m −2 .The National Climate Data Center Integrated Surface Hourly Database (Smith et al., 2011) has approximately 4000 stations that reported SunDu from 1982 to 2008.Data durations of about 2200 of these stations exceed one year, and about 1200 stations have more than ten-year data (Fig. 1).We also used SunDu data obtained from the China Meteorological Administration in this study.Most of the data are from the Northern Hemisphere.The Chinese and European datasets have the highest density.All the meterological data used in this paper are also from the Integrated Surface Hourly (ISH) Database released by the National Climate Data Center.

Method
The correlation between SunDu and R s was described by the Ångström formula (Kimball, 1919;Angstrom, 1924) that Prescott subsequently modified (Prescott, 1940).The modified expression assumes a linear relationship between relative R s and SunDu, and usually results in a good fit.However, the regression coefficients obtained may only apply to a particular location and not be applicable elsewhere (Sanchez-Lorenzo et al., 2009).Yang et al. (2006) proposed a physically-based hybrid model to estimate R s from SunDu.Their hybrid model followed the original Ångström-Prescott model but parameterized radiative extinctions of air and cloud separately.Daily solar radiation (R s ) can be parameterized as (Yang et al., 2006): where n is measured SunDu; N is the theoretical values of SunDu and R c is daily solar radiation under clear-sky conditions.N at a station depends on its location and day of a year.The effect of Rayleigh scattering, water vapor absorption, and ozone absorption, can be accounted for in R c by using meteorological observations including air temperature and humidity (Yang et al., 2006).The water vapor absorption of solar radiation has been included in the calculation of R c .This absorption contributed to the reported global dimming (about 1 W m −2 in 30 yr) as atmospheric water vapor content increased with global warming (Simmons et al., 2010).Yang et al. (2006) used winter-and summer-averaged aerosols based on Hess et al. (1998).The inter-annual variation of aerosols was not included.This inter-annual variation can be captured by the observed SunDu (Horseman et al., 2008;Kaiser and Qian, 2002;Raschke et al., 2006;Sanchez-Lorenzo et al., 2008;Sanchez-Lorenzo et al., 2009;Tang et al., 2011).Yang et al. (2006) used one year of SunDu data in Japan and R s to derive a 0 , a 1 , and a 2 in Eq. ( 1) and validated their values based on one year of USA data (1998), one year of Saudi Arabia data (1998), and two years of Chinese data (1997,1998).For this study, we used long-term measurements from a number of stations to improve parameter estimation.

Validation data
We used direct measurement of R s collected by GEBA (Fig. 1) to evaluate that derived from SunDu.Similarly to SunDu, the R s direct measurements from GEBA have the greatest density over Europe and China with European data quality possibly exceeding data quality over China (Fig. 1, see also Sect. 3.3).
The substantial variation of atmospheric aerosols over Europe and China, from the 1980s to the 2000s (Wang et al., 2009), provided a good test of whether SunDu measurements would accurately characterize these effects on R s .Therefore, we evaluated Eq. ( 1 1).More information on other stations is available in Table 1.
lected where GEBA R s and SunDu data overlap for at least 84 months from 1982 to 2008 (Table 1).
We used 80 stations, 40 in China and 40 in Europe (bold site name in Table 1), to derive the coefficients, and the other 57 stations to validate them.The derived coefficients of Eq. ( 1) are a 0 = 0.33, a 1 = 0.70, and a 2 = −0.02,and were used globally for all stations.Both European and Chinese sites were selected to derive and validate the coefficients to show the suitability of Eq. ( 1) for global usage.Table 1 confirms this by showing that Eq. ( 1) works well for both development and validation sites and it can be applied globally.Therefore, in the following discussion, we merged all the sites to show their regional variations.
The application of the method for estimating R s from SunDu depends on answers to the following questions: (a) Can SunDu accurately estimate the extent to which aerosols and cloud properties modify R s spatially and seasonally?(b) Can SunDu provide reliable estimates of interannual and decadal variations of R s ?The following two sections evaluate these two aspects.

Seasonal variation of R s
This analysis indicates that this method accurately predicts seasonal variations in R s .The seasonal variations are primarily determined by solar zenith angle and cloud variations.The overall standard deviation is 12 W m −2 (8.6 % in relative value) with a correlation coefficient of 0.98. Figure 2 shows a comparison between GEBA R s measurements and those derived from SunDu at one station.Table 1 summarizes the comparisons for all stations.
To further test our method, we calculated monthly anomalies in R s .For each station, we averaged all available R s data Table 1.A summary of the locations of the 137 stations in Europe and China where GEBA R s measurements and estimates from SunDu and aerosol optical depth (AOD) overlap for at least 84 months.The statistical parameters of the comparison between R s measurements and estimates are shown as standard deviation (STD), bias and averaged R s in W m −2 .To show the capability of the method to quantify the effect of clouds on R s , the statistical parameters of the comparisons of predicted and measured monthly R s anomalies (seasonal cycle removed) are also shown.Data at site names in bold were used to calibrate our method and others are used to validate the method.Table 1).More information on other stations is available in Table 1.
Fig. 3.An example of the comparison between monthly R s anomalies from direct measurements and our prediction from SunDu at GEBA station No. 1193 (Kucharovice, see Table 1).More information on other stations is available in Table 1.
during the study period from 1982 to 2005 for each month.
The monthly anomalies were obtained by removing the seasonal cycle from the original monthly R s data, separately for measured and predicted R s .Figure 3 illustrates one example of this comparison and all statistical parameters are shown in Table 1, with an overall correlation coefficient of 0.80 averaged from the statistical parameters from each station.
38 1 Fig. 4. Comparison between monthly regional Rs anomalies of measurements and our 2 prediction from SunDu.The regional average is over 86 stations in Europe (Table 1).Each 3 point represents one month.Comparison between monthly regional R s anomalies of measurements and our prediction from SunDu.The regional average is over 86 stations in Europe (Table 1).Each point represents one month.
We averaged the anomalies over Europe from all 86 stations, and compared the regional averaged anomalies to the GEBA directly measured values (Fig. 4), and over China from 51 stations (Fig. 5).SunDu-derived R s also predicts relatively well regional averaged anomalies in R s and it does a little better over Europe than over China based on regional monthly averages.39 1 Fig. 5. Comparison between monthly regional Rs anomalies of measurements and our 2 prediction from SunDu.The regional average is over 51 stations in China (Table 1).Each 3 point represents one month.Comparison between monthly regional R s anomalies of measurements and our prediction from SunDu.The regional average is over 51 stations in China (Table 1).Each point represents one month.
40 Fig. 6.Comparison of five-year smoothed monthly anomalies of SunDu-derived with directly-measured Rs averaged over 86 stations in Europe (Table 1).The breakpoint in 2003 reflects a break in data availability.The linear trends are calculated annual anomalies rather than five-year smoothed data.Fig. 6.Comparison of five-year smoothed monthly anomalies of SunDu-derived with directly-measured R s averaged over 86 stations in Europe (Table 1).The breakpoint in 2003 reflects a break in data availability.The linear trends are calculated annual anomalies rather than five-year smoothed data.

Decadal variation of R s
We also evaluated how well decadal variations in the SunDuderived R s are modeled.Figure 6 shows that the decadal variation in R s is adequately captured by SunDu-derived R s in Europe.Figure 7 shows that the SunDu-derived R s also reasonably predicts the variation before 1990 and after 1993 in China.However, there is a sharp increase in the measured R s between 1990 and 1993 that is not captured by SunDu. Figure 8 shows that SunDu-derived R s accurately predicts decadal variation in the baseline stations in China, where the pyranometers used to measure R s were carefully calibrated and maintained.The discontinuity in the other Chinese data (see Fig. 7) occurs during the period from 1990 to the spring of 1993, when China replaced its Russian-made pyranometers (Type: DFY) with Chinese-made pyranometers (Type: TBS) at most stations (CMA, 1996).
Therefore, it appears that the sharp increase in measured R s was primarily an artifact of instrument replacement and that SunDu-derived R s can be used to accurately predict decadal variations of R s in China.The decrease in SunDuderived R s in the period of 1990-1992 results from the abrupt increase in stratospheric aerosols (Stevermer et al., 2000) following the Pinatubo volcano eruption in 1991.Further evidence for this conclusion is provided by comparison of both estimates of R s with pan evaporation.Pan evaporation has been shown to be in good agreement with solar radiation (Roderick and Farquhar, 2002).The variation of pan evaporation from the 1980s to the 2000s in China (Cong et al., 2009) is more strongly correlated with SunDu-derived R s than the directly-measured R s shown in Fig. 7.The spuriousness of the discontinuity of R s direct measurements has also been confirmed by quantify-controlled process studies (Tang et al., 2010(Tang et al., , 2011)).
In summary, after converting SunDu into R s using the proposed method (Eq.1), SunDu-derived R s agrees very well with direct measurements on inter-annual and decadal time scales.

Inter-annual and decadal variability of regionally averaged R s
Since SunDu provides a good estimate of R s and is available in many regions, it can also be used to answer the following questions: (1) How has R s varied locally and regionally (i.e., 1982 to 2008); and (2) what have been the relative contributions from clouds versus aerosols to the observed variations of R s ?Decadal variations of R s around the globe are examined with data from stations that reported SunDu to the World Meteorological Organization, or in China from the Chinese Meteorological Administration.We selected 1165 stations where SunDu and other meteorological data are available for more than 120 months from 1982 to 2008 and divided the sites into six major regions according to their locations and long-term trends (Fig. 1).Daily R s were calculated and averaged into monthly values for comparison with measurements collected by GEBA.
Figure 9 shows that the R s derived from SunDu exhibits substantial decadal changes over most regions, similar to those displayed in previous studies (Gilgen et al., 1998;Liepert, 2002;Long et al., 2009;Ohmura, 2009;Stanhill and Cohen, 2001;Wielicki et al., 2002;Wild et al., 2005), and in particular, shows similar variations to those derived from the GEBA.In China, it was flat from the 1990s to 2000 but has been in a pronounced decline since 2000, substantiating the findings of a renewed dimming in China after 2000 (Wild et al., 2009;Xia, 2010).The sites that measure SunDu in China have a much higher density than those that give direct measurements of R s (Liang and Xia, 2005;Shi et al., 2008).Furthermore, these SunDu measurements indicate a substantially different decadal variation of R s over China from the direct measurements (cf., Figs.9b and 7). Figure 9b shows a substantial decadal variation of R s occurred in China and Fig. 9e shows that R s has decreased over Indonesia since 2004.

Clouds determine R s at a monthly time scale
We examined the contributions of clouds versus aerosols to the variations of R s .SunDu has been regarded as a direct measure of cloud cover fraction because all but the thinnest clouds, if intercepting the Sun, will decrease SunDu (Hoyt, 1977).Indeed, as shown in Fig. 10, relative SunDu (the ratio of measured SunDu to maximum possible SunDu during a day) on a monthly scale is highly correlated with ground measured total cloud fraction, i.e., cloud cover variability explains most of the SunDu, and hence R s variability on interannual or shorter time scales.

Aerosols contribute most of R s variability on a decadal time scale
Figure 9 correlates the SunDu-derived R s with independent data for aerosol optical depth (AOD) as well as cloud cover.
The total column AOD is estimated from tropospheric values (Wang et al., 2009) supplemented with satellite-measured stratospheric values (Stevermer et al., 2000).AERONET provides a better dataset of AOD than that derived from visibility (Holben et al., 1998).However, AERONET sites are very sparsely distributed and their data are of short duration.Long-term visibility data are available at most meteorological stations and they have been successfully used to determine long-term variation of tropospheric aerosols from 1973 to 2007 at more than globally 3000 stations (Wang et al., 2009).The clear-sky fraction is obtained from the total cloud cover fraction visually observed at weather stations.Such data from Canada and some countries in Europe have been omitted because of a change in methodology during the 1990's, i.e., surface-cloud observations were changed from human visual to instrumental assessment, resulting in a discontinuity in the data (Dai et al., 2006).Aerosols can only reduce SunDu at low solar elevations.Although, it has been questioned (Stanhill and Cohen, 2005) as to whether SunDu would detect the impact of aerosols on R s , existing studies have already shown that a reduction in SunDu correlates with increases in atmospheric aerosols (Horseman et al., 2008;Kaiser and Qian, 2002;Qian et al., 2006;Sanchez-Lorenzo et al., 2009).
Figure 9 shows that decadal variability of SunDu-derived R s is dominated by the changes in aerosols, especially those over Europe and China.The Pinatubo volcanic eruption in 1991 produced a large amount of stratospheric aerosols with substantial anomalies in global AOD and corresponded to a large reduction in R s in all regions (Fig. 9).The change of cloud cover in Europe compensated the impact of elevated stratospheric aerosols (Fig. 9a).Except during such eruptions, changes in R s on the decadal time scale are primarily determined by tropospheric AOD anomalies, e.g., the brightening over Europe after 1995, during which time Fig. 9a shows increasing cloud cover but decreasing AOD.Direct concurrent measurements of R s and AOD (Norris and Wild, 2009;Ruckstuhl et al., 2010) have confirmed that the AOD impact of R s , i.e., the solar irradiance increase caused by the direct effect of decreasing aerosol, accounts for most of the observed increase of all-sky solar radiation over Europe from 1981 to 2005.
The decline of R s since 2000 over China (Fig. 9b) was also a result of increased AOD, since cloud cover decreased at that time (Fig. 9b).The substantial increase of AOD in China from the late 1980s to the early 1990s, was compensated by a reduction in cloud cover (Qian et al., 2006;Warren et al., 2007;Xia, 2012) (Fig. 9b).The dips shown in Fig. 9e correspond to all the major fires occurring in Indonesia since 1990 (Podgorny et al., 2003;van der Werf et al., 2008van der Werf et al., ), in particular, the 1991van der Werf et al., , 1994van der Werf et al., , 1997van der Werf et al., -1998 fires, and the reported ramp up of burning after 2000 that culminated in the large fires of 2006 (Podgorny et al., 2003;van der Werf et al., 2008).
The connection between R s and AOD is further clarified by correlating their five-year smoothed monthly anomalies at each station, i.e., the month-to-month variability seen in Fig. 10 was filtered out. Figure 11 shows that about 58 % percent of all stations with this smoothing have a correlation coefficient larger than 0.5, indicating that the decadal variation in aerosols contributes more than 25 % of the decadal variance in R s at the majority of the individual stations.Some sites show low correlation or even negative correlation.At the stations, change of clouds is the determining factor of long-term variation of R s , such as those in the United States (Long et al., 2009).

Trend of R s over the Northern Hemisphere
We have merged measurements of R s derived by SunDu with those directly measured and collected by GEBA.Monthly anomalies were used to obtain a global trend.At each station, its significance was evaluated using the Mann-Kendall test, which is widely used for trend and changing point detection.About 44 % of the stations passed the 95 % confidence test.Figure 12 shows the long-term trends of R s as aggregated over 5 • × 5 • grid boxes (302 boxes).The 252 boxes obtained over the Northern Hemisphere have an average trend of 0.87 W m −2 per decade.Only 50 boxes were available for R s data over the Southern Hemisphere and these have a mean trend of zero.10).About 58 % of all stations have a correlation coefficient larger than 0.5.Such correlation coefficients indicate that the decadal variation in aerosols control the decadal variation in R s (Fig. 9).

Discussion and conclusions
SunDu is much more widely available and provides a longterm time series dataset where direct measurements are not available, such as in South Asia and South America.Akinoglu (2008) further pointed out that it is the only long term, reliable and readily available measured information that can be used to accurately estimate R s .Thus, estimation of R s by SunDu is useful as a complement to globally sparse direct measurements, even in Europe where direct measurements have the highest density.In this study, we investigated the variability of R s derived by SunDu from 2002SunDu from to 2008. .This study shows that SunDu data can describe variations of R s that are predominantly affected by either cloudiness or aerosols.The SunDu-derived R s agrees very well with direct measurements on inter-annual and decadal time scales.SunDu as a proxy for R s provides a stable long-term data series.Compared to its direct measurement, R s from SunDu appears to be less sensitive to instrument replacement and calibration, and shows that the widely reported sharp increase in R s during the early 1990s in China was a result of instrument replacement.
The estimates of R s by SunDu are merged with its direct measurements to provide a long global dataset with high spatial coverage over the Northern Hemisphere.This dataset shows an average increase at a rate of 0.87 W m −2 per decade from 1982 to 2008.The average trend over the Southern Hemisphere, whose coverage is much poorer, is negligible.Apparently, the long-term variations of R s over the two hemispheres are substantially different.The four major datasets used here, SunDu-derived R s , satellite-measured stratospheric AOD, visibility-derived tropospheric AOD, and visual observations of cloud cover fraction are totally independent.The consistency of their features show that the conventional SunDu observations supply robust information on atmospheric impact of the climate variability of R s .
As such, they would be useful to constrain climate model parameterizations that generate R s variability.Monthly variability of R s is determined by variability of cloud cover in most regions.The decadal variability of R s in Europe, China, and North America (primarily Canada and Mexico), is dominated by variations in tropospheric aerosols.
Further analysis would be needed to separate the contributions of direct, indirect, or semi-direct effects of aerosols from the effects of changes in cloud properties.To accomplish this, model simulation would be necessary.

Fig. 3 .
Fig. 3.An example of the comparison between monthly Rs anomalies from direct measurements and our prediction from SunDu at GEBA station No. 1193 (Kucharovice, See Fig. 4.Comparison between monthly regional R s anomalies of measurements and our prediction from SunDu.The regional average is over 86 stations in Europe (Table1).Each point represents one month.
Fig. 5.Comparison between monthly regional R s anomalies of measurements and our prediction from SunDu.The regional average is over 51 stations in China (Table1).Each point represents one month.

44Fig. 10 .
Fig. 10.Histogram of correlations between the relative SunDu (the ratio of measured SunDu to theoretical SunDu during a day) and monthly anomalies of total cloud cover fraction.Total cloud cover fraction is observed visually by trained technicians at weather stations.The plot shows that relative SunDu is highly correlated with cloud cover fraction on a monthly scale at most stations.

Fig. 10 .
Fig. 10.Histogram of correlations between the relative SunDu (the ratio of measured SunDu to theoretical SunDu during a day) and monthly anomalies of total cloud cover fraction.Total cloud cover fraction is observed visually by trained technicians at weather stations.The plot shows that relative SunDu is highly correlated with cloud cover fraction on a monthly scale at most stations.

Fig. 11 .
Fig. 11.The histogram of correlations between the five-year smoothed Rs and -AOD monthly anomalies at each station.The five-year (60-month) smoothing filters out high frequency variations in Rs and -AOD (shown in Fig.9).About 58% of all stations have a correlation coefficient larger than 0.5.Such correlation coefficients indicate that the decadal variation in aerosols control the decadal variation in Rs (Fig.9).

Fig. 11 .
Fig. 11.The histogram of correlations between the five-year smoothed R s and -AOD monthly anomalies at each station.The five-year (60-month) smoothing filters out high frequency variations in R s and -AOD (shown in Fig.10).About 58 % of all stations have a correlation coefficient larger than 0.5.Such correlation coefficients indicate that the decadal variation in aerosols control the decadal variation in R s (Fig.9).

Fig. 12 .
Fig. 12.The linear trend of Rs from 1982 to 2008 (unit: Wm -2 per decade) from merged 2 SunDu-derived values and direct measurements collected by GEBA.The trends are first 3 calculated at each station and then aggregated into a 5°×5° grid and the averages are shown.4 There are 302 such grids obtained and shown.5 6 ) in Europe and China from 1982 to 2005 when both SunDu and GEBA station data were available.86stationsoverEuropeand 51 stations over China were se-An example of the comparison between monthly averages of solar radiation (Rs) 2 measurements and our prediction from SunDu at GEBA station No. 1193 (Kucharovice, 3 CZECH REPUBLIC, See Table1).More information on other stations is available in Table1.An example of the comparison between monthly averages of solar radiation (R s ) measurements and our prediction from SunDu at GEBA station No. 1193 (Kucharovice, CZECH REPUBLIC, see Table