Reconstruction of erythemal UV-levels for two stations in Austria: a comparison between alpine and urban regions

The aim of this study is the reconstruction of past UV-radiation doses for two stations in Austria, Hoher Sonnblick and Vienna, using a physical radiation transfer model. The method uses the modeled UV-radiation under clear-sky conditions, cloud modiﬁcation factors and a correction factor as input variables. To identify the inﬂuence of temporal 5 resolution of input data and modiﬁcation factors, an ensemble of four di ﬀ erent modelling approaches has been calculated, each with hourly or daily resolution. This is especially important because we found no other study describing the inﬂuence of the temporal resolution of input data on model performance. Following the results of the statistical analysis of the evaluation period the model with the highest temporal resolution has 10 been chosen for the reconstruction of the UV-radiation doses. This model (HMC) uses modelled UV-radiation under clear sky conditions, a cloud modiﬁcation factor, both with hourly resolution, and a monthly correction factor. A good agreement between modelled and measured values of erythemally e ﬀ ective irradiance was found at both stations. In relation to the reference period 1976–1985 an increase in erythemal UV- 15 irradiance in Vienna of 11 percent is visible in the period 1986–1995 and an increase of 17 percent in the period 1996–2005 can be seen. At Hoher Sonnblick an increase of 2 percent has been calculated for the yearly averages in erythemal UV for the period 1986–1995 and an increase of 9 percent for the period 1996–2005 in comparison to the reference period. For the di ﬀ erent seasons the strongest increase in erythemal UV 20 radiation has been found for winter and spring season at both stations. correlation between observed and modelled UV irradiance for the testing period using the HMC-Model for the di ﬀ erent seasons. The results show a good agreement throughout the year. The correlation coe ﬃ - 25 cient for all seasons is between 0.96 and 0.98 at Hoher Sonnblick and 0.96 and 0.99 at Vienna. The Bias is between − 8 and + 1 percent at Hoher Sonnblick and between − 1 and + 7 percent in Vienna. The values of the root mean square error are between + 18

EGU global ozone concentration of about 10 percent would lead to an increase in skin cancer cases of 300 000 per year (IPCS, 1994). Climate Change may also have an effect on future UV radiation doses reaching the earth's surface through changes in cloud amount, cloud properties and surface albedo (e.g. Lindfors and Vuilleumier, 2005). Increased research activities in the field of UV radiation have led in the last years 5 to worldwide monitoring of UV radiation levels (e.g. Seckmeyer et al., 1995), a better understanding of the atmospheric processes that influence the UV part of the spectra (e.g. Weihs and Webb, 1997a, b;Schwander et al., 1997) and the development of modelling approaches for the reconstruction of past UV radiation levels (e.g. Lindfors and Vuilleumier, 2005;Reuder and Koepke, 2005;Kaurola et al., 2000;Gantner et al., 10 2000).
Recent studies have shown that models may be able to reconstruct past UV levels with an accuracy between 5 and 15 percent (Fioletov et al., 2001;Kaurola et al., 2000;Reuder and Koepke, 2005). Most of them were based on daily values. A few studies however already use hourly meteorological input data (Reuder and Koepke, 2005; Den 15 Outer et al., 2005). The exact improvement in accuracy by using hourly data has however not really been investigated. Only a few studies have given a statement on the exact magnitude of the meteorological factors influencing reconstructed UV levels (e.g. Lindfors and Vuilleumier, 2005;Reuder and Koepke, 2005).
During the last decade reconstruction studies for several regions in Europe, espe-20 cially in Nordic and central European countries, have been performed (e.g. Lindfors et al., 2003;Lindfors and Vuilleumier, 2005;Reuder and Koepke, 2005;Den Outer et al., 2005). To the knowledge of the authors no study on the reconstruction of UV levels in a high alpine region above 3000 m has been performed. Lindfors and Vuilleumier (2005) reconstructed erythemal UV-radiation for the alpine station Davos, which is located on 25 an altitude of 1850 m a.s.l., but this is still 1256 m lower than the station Sonnblick used in this study. During the present study the first reconstruction for an alpine site, located at more than 3000 m. a.s.l, was performed. The unique location of the mountain observatory at Hoher Sonnblick in Europe provides an opportunity to reconstruct solar UV EGU for a high altitude alpine site and to investigate the influence of topography connected with high ground albedo and very low atmospheric pollution levels on UV trends. The present study investigates first the influence of different temporal resolutions of input data on the reconstruction accuracy, it then performs a reconstruction for one high altitude alpine site and one low altitude site and investigates the contribution of 5 the different meteorological parameters on the changes in UV. The investigation of changes in UV for the low altitude site of Vienna is especially of interest because this region is, with its 1.6 million inhabitants, the most populated one in Austria. These are to the knowledge of the authors new aspects in the domain of UV reconstruction. 10 The Austrian national UV Monitoring network has been in operation since 1998 and has 12 stations all over the country. As already mentioned two stations out of this network have been selected to reconstruct past UV-levels. One station the Observatory Sonnblick (47 • 03 ′ N,12 • 57 ′ E) is located in the central alps at an altitude of 3106 m. The other one, Vienna (48 • 15 ′ N,16 • 26 ′ E), is located in the north-eastern part of the 15 country at an altitude of 153 m. These two stations have been selected because of their significant differences in altitude, terrain and meteorological conditions. The data used in this study include UV erythemally weighted broadband irradiance, global irradiance, sunshine duration, snow height and snow cover as well as total ozone column.

20
In 1996, the setup of an Austrian UVB monitoring network was initiated by the Federal Department of Environment (Blumthaler and Schauberger, 2001

EGU
The relative spectral response of each detector is determined and by comparison with a double monochromator spectroradiometer the absolute calibration function is derived in dependence on solar zenith angle and on total atmospheric ozone (Blumthaler, 2004). The uncertainty of the calibration is about ±7% (at 95% confidence level) for solar zenith angles <75 • , which is dominated by the uncertainty of the calibration lamp for the 5 spectroradiometer (±4%). During routine operation, the measurements of all detectors are transmitted in near real time to the laboratory and then converted to UV-Indices, the internationally agreed unit for erythemally weighted solar irradiance. The results are then published on the internet (http://www.uv-index.at) every 15 min., together with a regional map showing the distribution of the UV-Index over Austria by combining 10 the information from the measurement detectors with cloud information from Meteosat Second Generation.

Total ozone
The total ozone content, usually given in Dobson Units (DU), is one of the most impor-15 tant parameters affecting UV-B radiation at the ground. Ground based measurements by Dobson (e.g. Dobson, 1931;Komyhr, 1980) and Brewer (e.g. Brewer, 1973;Kerr and McElroy, 1995) instruments are the most accurate and widely used method for the determination of atmospheric total ozone content. Because of the lack in local ozone data before 1994 at Hoher Sonnblick and the absence of ozone measurements 20 in Vienna additional datasets from satellite, ground measurements as well as ozone model simulations have been used. The total ozone record from Arosa is the world's longest, dating back to 1926. This record has been homogenized and is discussed in detail by Staehelin et al. (1998a, b is assured by a comparison of the modeled total ozone with the ground-based data taken at several Dobson stations which have been in operation since the early 1950s and 1960s. A comparison against the measured daily total ozone over Arosa provides that the mean difference, [(Model -Observation)/Observation]*100%, is 1.5%±4.5% for the 10 period 1976-2004. A high correspondence between the modeled and measured total ozone values over Arosa is also corroborated by Fig. 1a. The model estimates total ozone values at sea level pressure, thus an equivalent of measured total ozone at Arosa (site altitude ∼1600 m) should be lowered of ∼1.5% because of the altitude effects, i.e. ∼1%/km decrease of total ozone is approximated from an assumption that 15 ∼10% of column ozone content is in the troposphere.
We compared the modeled daily total ozone values for Arosa and Vienna (or Hoher Sonnblick), to find a relation between the ozone values for these sites. The mean difference, [(Model Vienna -Observation Arosa)/Observation Arosa]*100%, is 4.0%±6.5% and the correlation coefficient between the modeled values for Vienna and the ob-20 served ozone at Arosa is 0.92 (see also Fig. 1b). Moreover, the mean difference between modeled daily ozone values for Vienna and Arosa (the altitude effect is not included) is 2.5% for the period 1976-2004. Thus, the measured Arosa total ozone should be enlarged by 4.0% to be representative for Vienna. Similar calculations done for Hoher Sonnblick show that measured total ozone for Arosa corresponds to 25 total ozone that would be expected over Hoher Sonnblick at altitude 3100 m (see also Fig. 1c) for a correspondence of modeled total ozone values for these sites). Moreover, Introduction EGU the comparison of the measured monthly averages between the Austrian ozone data set and those from Arosa shows that the differences are not higher than ±3%. We may therefore use the measured total ozone at Arosa to cover the whole period for Austria. The ozone record of Arosa has some periods of missing data (see Staehelin et al., 1998a, b) and for these periods missing values are filled up by the modeled COST-726 5 total ozone.

Cloudiness, global irradiance and sunshine duration
The use of cloud modification factors (CMFs), gained from information on global irradiance and sunshine duration, is a common method used by the scientific community to describe cloud effects on solar UV radiation (e.g. Kaurola et al., 2000;Schwander et 10 al., 2002;Koepke et al., 2006). European Union's Action COST 726 identified global irradiance in combination with cloud modification factors (CMFs) as the best way to describe cloud attenuation and cloud effects on solar UV (Koepke et al., 2006). We therefore used global irradiance data and sunshine duration only if global irradiance data were missing. Global irra-15 diance and sunshine duration have been measured at both stations by the Austrian Central Institute for Meteorology and Geodynamics (ZAMG). The time series for global irradiance and sunshine duration at Hoher Sonnblick is dating back to the year 1972 while at Vienna information is available since the year 1960. The global irradiance time series has some gaps at station Hoher Sonnblick (see Table 1). Sunshine duration 20 was therefore used to fill the gaps. Sunshine duration was used to distinguish between clear-sky and cloudy conditions and to reconstruct global irradiance levels for some periods of the used time series. More information will be given in Sect

EGU
3.3 Ground albedo, snow depth and snow amount Snow depth and snow amount are the most important factors influencing surface albedo in the UV range of the spectra and the relationship has been well studied (Blumthaler and Ambach, 1988;McKenzie et al., 1998;Weihs et al., 1999;Kylling et al., 2000;Weihs et al., 2001;Schmucki et al., 2002). The data on snow depth and 5 snow amount used in our study were measured by the Austrian Central Institute for Meteorology and Geodynamics (ZAMG). At Hoher Sonnblick measurements on snow depth and snow amount are available since 1972 while in Vienna information is available since 1960. Snow cover is the most important parameter for the calculations of the surface albedo (see Sect. 4.1).

Methods
As already mentioned in Sect. 3.2., we used information on global irradiance instead of sunshine duration to calculate past UV levels. For the reconstruction an approach similar to the one described by Kaurola et al. (2000) has been used. The use of cloud modification factors (CMFs) to model erythemal UV irradiance for skies with broken cloudi-15 ness in combination with a clear-sky model is a commonly used method (e.g. Thiel et al., 1997;Schwander et al., 2002;Koepke et al., 2006;Kaurola et al., 2000). Kaurola et al. (2000) used modelled clear-sky UV values, a cloud modification factor for global irradiance and a correction factor in their study (Eq. 1). Here UV REKO is the reconstructed UV-radiation, UV MOD is the modelled UV-radiation under clear-sky 20 conditions, CMF SOL is the cloud modification factor for global irradiance and C is a seasonal correction factor.
For our purpose we have modified this approach for different temporal resolutions which are discussed in detail in Sect. 4.4 of this paper.
Albedo=0.138 + 6.99 * 10 −4 * SNOW + 6.06 * 10 −3 * NSNOW−1.489 * 10 −2 * DAYS (3) 10 In Vienna snow cover is less frequent and we used fixed ground albedo values for snow covered and snow free ground conditions. If a snow height of more than 5 mm was reported in Vienna we used a value of 0.4, if the snow cover was below this value we used a value of 0.03 to describe the surface albedo.

15
To model clear-sky UV radiation the model SDISORT developed by Stamnes et al. (1988) has been used. This model needs information on date, time, location, altitude, solar zenith angle, total ozone column, surface albedo and aerosol optical depth as input parameters. Using SDISORT we calculated clear-sky UV levels on hourly resolution (UV MODEL ).  1973, 1981, 1983, 1984, 1988 and 1990. Global irradiance is needed to calculate the cloud modification factors for global irradiance. We modified an approach developed by Neuwirth (1979) for the reconstruction of monthly mean values of global 5 irradiance (see Eqs. 4 and 5), whereas G OBS is the observed global irradiance, G POT is the potential global irradiance, a and b are correction factors, n is the observed sunshine duration and N is the potential maximum sunshine duration.
For our study we adapted the model for hourly values (Eq. 6). Where G MODEL is the modelled global irradiance, G REF is the reference global irradiance under clear-sky conditions, X is a correction factor dependent on observed sunshine duration and N is the potential sunshine duration.
Using this modelling approach 90% of the modelled global irradiance values agree within ±20% of the observed ones.

Calculation of cloud modification and correction factors
The cloud modification factor for global irradiance (CMF SOL ) is calculated as shown in Eq. (7) as ratio between observed (G OBS ) and reference global irradiance (G REF ) for 20 each solar zenith angle (za).

EGU
Corresponding to the cloud modification factors for global irradiance a CMF for the UV part of the spectra (CMF UV ) has been calculated (Eq. 8). The CMF UV is calculated as ratio between observed UV radiation (UV OBS ) and modelled clear-sky UV radiation (UV MOD ) for each solar zenith angle (za). The CMF values were stored in look up tables for a solar zenith angle range between 19 • and 90 • using 2 degree intervals.
The correction factor C has been calculated as ratio between CMF SOL and CMF UV (see Eq. 9).
4.5 Analysis of the influence of different temporal resolutions on the model perfor- To analyse the influence of temporal resolutions of input data, cloud modification and correction factors, 4 modelling approaches with different temporal resolution were developed (Eqs. 10 to 13). EGU global irradiance on hourly resolution and CMF SOL(D) an average daily cloud modification factor for global irradiance, C M is a monthly correction factor and C S a seasonal correction factor. For each station we defined two independent data sets: a 2 year development period and a 2 year testing period. The performance of our 4 models is quiet different 5 throughout the year. Table 2 shows the correlation, root mean square error and bias between estimated and observed daily UV doses for the different seasons of the year for all 4 modelling approaches. At station Hoher Sonnblick the HMC-Model shows the best fit between estimated and observed UV doses for all 4 seasons. In Vienna the HMC-Model shows the best fit during the winter, spring and summer seasons while in 10 autumn the reconstruction quality of the HMC, HSC and DMC are similar.
A more quantitative way to look on the performance of our reconstruction approaches for both stations can be seen in Table 3 (for Vienna) and Table 4 (for Hoher Sonnblick). It is obvious that the reconstruction quality significantly decreases with increasing temporal resolution of the input data. Altogether, most of the day-to-day variation is cap-15 tured fairly well by our models and the results are comparable to those from other studies (Lindfors and Vuilleumier, 2005;Reuder and Koepke, 2005;Kaurola et al., 2000). The performance of the single models on monthly timescale is even better. The results in Tables 3 and 4 also point out that the performance of our reconstruction method is slightly better at Hoher Sonnblick than for Vienna. Possible reasons for that will be 20 discussed later in this paper.

EGU
and +11 percent at Hoher Sonnblick and between +26 and +13 percent in Vienna.

Results
The results of the statistical analysis of the testing period (Sect. 4.5) showed that the model with the highest temporal resolution (HMC) shows the best agreement between observed and modelled UV values. Using the method described above, the hourly and 5 daily erythemal UV doses at Hoher Sonnblick and Vienna were estimated for the last decades. The following comparison concentrates on the period 1976-2005 in order to have a comparable 30 year period of data at both stations. Figure 4 shows the yearly averages of estimated UV doses compared with the stratospheric total ozone content since 1976 for both stations. In comparison to the reference 10 period 1976-1985 we found an increase in the yearly averages of erythemal UV radiation for the period 1986-1995 of 11% in Vienna and 2% at Hoher Sonnblick and for the period 1996-2005 of 17% in Vienna and 9% at Hoher Sonnblick. Figure 5 shows the yearly averages of estimated UV doses and relative sunshine duration since 1976 for both stations. The results of the calculated averages of erythemal UV radiation and 15 atmospheric total ozone content are plotted in Fig. 6a-d for each season separately. The plots of the seasonal averages of erythemal UV radiation compared to the relative sunshine duration are shown in Fig. 7a-d.

Total ozone column
Compared with the reference period 1976-1985 we found for the total ozone column 20 during the period 1986-1995 an average reduction of 5% for winter, 4% for spring and 2% in summer and autumn. For the period 1996-2005 an average reduction in the ozone content of 4% for the winter months, 6% for spring, 4% for summer and only 1% for autumn was found. For the yearly averages we found a reduction in total ozone of 4% for the period 1986-1995 and of 5% for the period 1996-2005  EGU months and 10% in autumn. Surprisingly there is no change during the summer within both periods related to the reference time.

Influence of ozone and sunshine duration on erythemal UV-radiation
To identify the influence of total ozone content and sunshine duration on the erythemal UV-radiation doses we performed model simulations holding atmospheric ozone 5 content and sunshine duration fixed at 1960s levels. We calculated the influence of changes in total ozone content on erythemal UV following Eq. (14) and the influence of changes in cloud cover following Eq. (15).

M TOC= UV KSD
UV REKO * 100 (14) M SD= UV KTOC UV REKO * 100 (15) 10 Where M TOC is the change in erythemal UV through the influence of total ozone, M SD is the change in erythemal UV through the influence of changes in cloudiness, UV REKO is the reconstructed UV under observed conditions, UV KSD is reconstructed UV holding cloudiness constant on 1960s levels, UV KTOC is reconstructed UV holding total ozone concentration fixed on 1960s levels.

15
Using this technique we found out that changes in the atmospheric total ozone content play in most cases the dominant role in influencing erythemal UV-radiation changes at both stations (Tables 6 and 7). In Vienna during the period 1986-1995 for winter, summer and autumn more than 60% of the changes in erythemal UV result from changes in atmospheric ozone concentration. During spring season about 88% of 20 the changes in erythemal UV result from the decrease in stratospheric ozone content. For the period 1996-2005 we found similar results. Here about 60% of the changes in erythemal UV arise from changes in total ozone concentration during spring, summer and autumn. During winter we found that 70% of the changes in erythemal UV result 972 Introduction EGU from changes in total ozone and only 30% arise because of the increase in relative sunshine duration. At Station Hoher Sonnblick during the period 1986-1995 an equal influence of changes in stratospheric ozone content and sunshine duration during winter and summer was found. Half of this increase in erythemal UV occurs through a decrease in total ozone content, the other one through the increase in sunshine du-5 ration. During springtime the decrease in stratospheric ozone is the major influencing factor on erythemal UV-radiation. The small decrease in sunshine duration shows almost no effect on the erythemal UV dose. In autumn the small decrease in erythemal UV-radiation results only from decrease in relative sunshine duration. In the period 1996-2005 during winter, 60% of the changes in erythemal UV-irradiance result from 10 decreasing stratospheric ozone concentrations. For spring and summer, change in the erythemal UV is due to changes in cloudiness and changes in stratospheric ozone content. For autumn the control run shows a decrease in erythemal UV resulting from the decrease in cloudiness. Because of that the decrease in total ozone concentration is the main cause for the overall increase in erythemal UV. The results for the different 15 periods of the year are summarized for both stations in Table 6. About 66% of the changes of the yearly averages of the erythemal UV radiation results from changes in the total ozone concentration, whereas 34% are caused through changes in cloudiness. The changes in yearly averages of erythemal UV at Hoher Sonnblick during the period 1986-1995 are partly caused by decreasing atmospheric ozone concentra-20 tions and partly by changes in cloudiness. For the period 1996-2005 we found that the major influence on the average change in erythemal UV was total ozone. Here 78% of the changes in erythemal UV are caused by decreasing atmospheric ozone concentrations. These results are summarized in Table 7.
6 Summary and discussion 25 In our study we performed the first reconstruction for erythemal UV irradiance in Austria for one alpine and one urban region. A method developed by Kaurola et al. (2000) ACPD Introduction EGU was adjusted to our meteorological stations. The method, which is fairly simple uses modelled clear-sky UV-radiation, cloud modification factors and correction factors and enables the estimation of UV radiation at Hoher Sonnblick and in Vienna for the last decades. For the modelling of clear-sky UV radiation we used SDISORT developed by Stamnes et al. (1988), which needs information on date, time, location, altitude, 5 solar zenith angle, ground albedo, total ozone column and optical thickness of the atmosphere as input parameters. Further we analysed the influence of temporal resolution of input data, cloud modification factors and correction factors on the model performance. The improvement in modelling accuracy by changing the time resolution could therefore be estimated. The results of the testing period show clearly that those 10 modelling approach using input data and modification factors with the highest temporal resolution shows the best fit between estimated and observed UV doses. EGU snow cover in the UV part of the spectrum in Vienna. In addition aerosol loading was assumed constant during the period of our study. Since aerosol concentration is much higher in Vienna than at Sonnblick it may have larger effects on calculation uncertainty in Vienna than at Sonnblick. The measured CMF in the short wave length range may partly take into account the turbidity but it 5 may be insufficient to characterize the full aerosol effects in the UV. But aerosol optical thickness has fortunately not increased much in central Europe since 1980 (Tegen et al., 2000). Another source of uncertainty was the lack of total ozone data in Vienna and at Hoher Sonnblick before the year 1994. Alternative ozone data from Arosa (CH), satellite data and ozone data obtained from modelling (see Sect. 3.1.) had therefore 10 to be used. We compared the monthly means of total ozone concentration between Hoher Sonnblick and Arosa and there we found that the difference in the period 1994-1998 did not exceed ±3 percent. Through the results of a careful statistical analysis we finally believe that the alternative ozone dataset used in periods with missing total ozone observations over Arosa should not be a major source of uncertainty because 15 on the one hand atmospheric ozone concentrations are spatially quite homogeneous and on the other hand we developed our correction factors with data obtained with the alternative dataset of total ozone (COST-726 total ozone data base providing high quality daily ozone data covering whole Europe since January 1950, Krzyścin, 2007 and adjusted so our modelled UV doses. 20 The analysis of our reconstructed time series shows a clear signal of increasing UV radiation in the last two decades in Austria. A higher increase in UV was calculated for the winter and spring season, and this is explained by a larger decrease in atmospheric ozone concentration in combination with an increase in sunshine duration. The results from our control run show that generally changes in atmospheric ozone concentrations 25 have influenced erythemal UV more than changes in cloudiness. This study provides important information for epidemiological studies and for future required medical care in Austria, knowing that erythemal UV radiation is one of the major causes for development of skin cancer.