**8. Conclusion**

48 Solar Radiation

manifesting also overall negative trend. During the last two decades, on the average, the ratios Mean/Trimean were larger, and the values of all three central trend measures of

> 1950 1960 1970 1980 1990 2000 2010 Year

> 1950 1960 1970 1980 1990 2000 2010 Year

Fig. 13. Switching from brightening to dimming trend in October

G/Gclear were smaller than before.

0.0

0.0

0.2

0.4

Trimean of monthly values (October)

0.6

0.8

Fig. 12. Trend of dimming for January in 1963-2008

0.2

0.4

Trimean of monthly values (January)

0.6

0.8

The relative availability of solar radiation at moderate and subpolar latitudes manifests no significant correlation between seasons. The weather regimes often change significantly during each of the conventional four seasons. The annual available amount of solar energy at these latitudes depends on the contributions of spring and summer seasons. For solar radiation applications study of the interannual variations and trends on the seasonal and even on the monthly level is necessary. Detailed data on interannual variation of the available solar energy within seasonal and often even monthly limits are useful for estimation of the biospheric responses, local potential for recreation services as well as for food and clean energy production.

The results obtained at one Estonian site could be considered as an example manifesting that the variations and trends of the relative availability of solar radiation may be substantially different in different seasons and months. Statistical analysis of trends and interannual variations needs at first finding the most appropriate general trend measures based on the probability density distributions of daily amounts of relative global irradiance within selected time intervals. In the case of symmetric distributions the conventional mean is a good measure but in the case of skewed distributions using of it leads to distorted interseasonal proportions and trends.

Using conventional mean instead of more robust trimean, leads to overestimation of the monthly general trend of G/Gclear in October to February by 4.5 % to 14% on average, depending on the month. In cases of thick clouds domination in separate years, the values of monthly mean may be by 30-40 % larger than trimean and median. In some cases the values of conventional mean were found to be up to 10 % lower than the trimean and median. In bright half-year, April to August, the conventional mean was consistently about 2.5 % lower than the trimean, and using it leads to some underestimation of central tendency in G/Gclear . Using conventional mean as a measure of monthly central tendencies is related to apparent reduction of the contrast between relative solar energy supply in summer and winter.

Monthly relative availabilities of broadband solar irradiance at the study site and presumably also in its wider neighbourhood varied in wide ranges. The ranges of variations were smaller in months of summer half-year and larger in autumnal and winter seasons. The average availability of global irradinance in summer half-year, and separately in spring and summer seasons, was 0.65 and that of direct irradiance 0.41 of the assumed normal cloudless weather amount. In spring the relative availability of direct irradiance has been to some extent larger than in summer and its interannual variation smaller.

Slopes of the linear trends were found to be smallest in the case of using conventional mean as a central tendency measure. It leads to underestimation of the dimming and brightening trends in the cases when linear approximation is appropriate for their description. In summer months, in most cases, small brightening trends were found. In June it changed to rather a dimming trend around 1973. In winter months there appears an overall dimming tendency of G/Gclear in recent two decades. Using the conventional mean leads to underestimation of that trend by up to 20-30 % as compared to the version calculated on the basis of trimean.
