Comparison of UV climates at Summit,

An SUV-150B spectroradiometer for measuring solar ultraviolet (UV) irradiance was installed at Summit, Greenland, in August 2004. Here we compare the initial data from this new location with similar measurements from Barrow, Alaska and South Pole. Measurements of irradiance at 345 nm performed at equivalent solar zenith angles (SZAs) are almost identical at Summit and South Pole. The good agreement can be explained with the similar location of the two sites on high-altitude ice caps with high surface albedo. Clouds have little impact at both sites, but can reduce irradiance at Barrow by more than 75%. Clear-sky measurements at Barrow are smaller than at Summit by 14% in spring and 36% in summer, mostly due to differences in surface albedo and altitude. Comparisons with model calculations indicate that aerosols can reduce clear-sky irradiance at 345 nm by 4?6%; aerosol influence is largest in April. Differences in total ozone at the three sites have a large influence on the UV Index. At South Pole, the UV Index is on average 20?80% larger during the ozone hole period than between January and March. At Summit, total ozone peaks in April and UV Indices in spring are on average 10?25% smaller than in the summer. Maximum UV Indices ever observed at Summit and South Pole are 6.7 and 4.0, respectively. The larger value at Summit is due to the site's lower latitude. For comparable SZAs, average UV Indices measured during October and November at South Pole are 1.9?2.4 times larger than measurements during March and April at Summit. Average UV Indices at Summit are over 50% greater than at Barrow because of the larger cloud influence at Barrow.

and the long hours of sunshine during summer months can lead to considerable UV exposure (Cockell et al., 2001). On the other hand, the virtual absence of UV-B radiation during winter months may result in Vitamin D deficiency (Webb et al., 1988;Engelsen et al., 2005) and associated diseases such as rickets (Stokstad, 2003).
Stratospheric ozone depletion and climate change (in particular changes in strato- 10 spheric temperatures, surface albedo, cloudiness and atmospheric circulation patterns) may modify the UV climate in the Arctic (Bernhard et al., 2007). Ozone decreases in the Arctic are less severe than in the Antarctic because of higher stratospheric temperatures in the North. Increased temperatures make the formation of polar stratospheric clouds (PSC) and the photochemical destruction of ozone less prevalent (WMO, 2007). 15 However, in years when the polar vortex was strong and stratospheric temperatures remained cold (e.g. 1993, 1997, 2005), photochemically induced decreases in total ozone of up to 45% have been observed over localized areas of the Arctic (Fioletov et al., 1997;Newman et al., 1997). Stratospheric temperatures during the winter/spring season are currently often close to the threshold of PSC formation. Reductions of 20 stratospheric temperatures as a consequence of climate change, could increase the frequency of years with severe ozone depletion in the Arctic (Rex et al., 2004;WMO 2007) and lead to marked increases in UV. Several studies have shown a longitudinal component of Arctic ozone concentrations, which have been explained by preferential locations of the Northern Polar Vortex, and by decadal variations in circulation patterns 25 (Knudsen and Andersen, 2001;Andersen and Knudsen, 2006). It can therefore be assumed that changes of UV radiation will not be uniform across the Arctic.
Measurements of solar UV radiation in the Arctic are sparse despite the potentially large changes anticipated in the future (Taalas et al., 2000). Current (Bernhard et al., 2007), Canada (Fioletov et al., 2001), Scandinavia (Lakkala et al., 2003), and Spitzbergen (Wuttke et al., 2005). There are no measurements in the vast region of the Russian Arctic. Here we report on first results from a new UV monitoring site, which was established in August 2004 at "Summit," located at the top of Greenland's ice cap. The instruments located at South Pole Station, Antarctica (90˚S, 2841 m a.s.l.), and Barrow, Alaska (71 • 19 ′ N, 156 • 41 ′ W, 8 m a.s.l.), are SUV-100 spectroradiometers (Booth et al., 1994), and were installed in 1988 and 1991, respectively. The instruments provide similar data products than the SUV-150B but have a larger bandwidth of 1.0 nm FWHM. 5 South Pole Station (SPO) and Summit (SUM) are located on the top of vast ice caps with a surface albedo of larger than 0.97 year-round. Barrow (BAR) is located at the coast of the Chukchi Sea, which is typically covered by ice between November and July. Snow cover extends roughly from October to June. The effective surface albedo is 0.83±0.08 (±1σ) during March and April and below 0.15 during August and September. 10 Clouds are frequent in any month but have a greater optical depth in autumn than in spring. More information on data from SPO and BAR has been compiled by Bernhard et al. (2004;. Data from SUM presented here cover the period of August 2004-July 2007 (42 769 spectra). There is a gap between 18-May 2005 and 1-August 2005 when the system 15 had to be removed for relocation of the host building. Data from SPO are from January 1991-January 2007 (182 495 spectra), and BAR data include January 1991-November 2006 (193 026 spectra).

Data analysis
All data used in this study are "Version 2" NSF network data (Bernhard et al., 2004), 20 which have been corrected for the instruments' wavelength and cosine errors, and adjusted for drifts of responsivity over time. All data are available via the web site www.biospherical.com/NSF/Version2. While the correction procedures are identical for all instruments, their magnitudes are different due to the different specifications of the SUV-150B installed at SUM and the SUV-100 spectroradiometers operating at SPO 25 and BAR. The maximum cosine correction of SUV-150B data in the UV is 3% while the correction for SUV-100 data may be as large as 16%. Corrections for wavelength er- Interactive Discussion rors are also smaller for the SUV-150B due its advanced drive using optical encoders. SUV-150B measurements were initially affected by large changes in responsivity of up to 10% per month, however. These changes were mostly due to a rapid degradation ("yellowing") of the Barium-sulfate coating of the integrating sphere used in the instrument's original cosine collector. In response to the problem, the integrating sphere was 5 replaced in August 2005 with one made from solid sections of PTFE. The replacement reduced the instrument's instability considerably to less than 2% per month. These drifts were monotonic and predictable over time, and could be corrected by linear interpolation of calibration scans, which were performed every two weeks.
To confirm that changes in responsivity have been adjusted appropriately, we com-10 pared SUV-150B data with measurements of a co-located GUV-511 multi-channel filter radiometer. For this comparison, SUV-150B spectra were weighted with the measured response functions of the GUV-511 as described by Bernhard et al. (2005). The analysis is complicated by the fact that the responsivity of the GUV-511 also changed over time. Between May 2004 and July 2007, the sensitivity of the GUV decreased mono-15 tonically by 19% at 320 nm, 7% at 340 nm, and 10% at 380 nm. To decouple the drifts of the two instruments, we calibrated the GUV-511 radiometer against SUV-150B data using synchronous measurements collected during periods ranging from three to five months. Drifts of the GUV's 340 and 380 nm channels during these periods are smaller than 1.5%. The method blends the strengths of both instruments: the respon-20 sivity of the SUV-150B is accurately known at the times of the bi-weekly calibrations, and the comparatively good short-term stability of the GUV-511 allows the assessment of changes in SUV-150B responsivity between calibration events. Figure 1 shows the ratio of SUV-150B and GUV-511 measurements for 6 different periods. The ratios show discontinuities of up to 4% at times when the calibration of the 25 SUV-150B was changed, but the overall agreement is good: standard deviations of the ratios of SUV/GUV range between 0.011 and 0.022 for all periods and wavelengths. The average standard deviation is 0.015, confirming that responsivity changes of the SUV-150B have been corrected satisfactorily.

Uncertainty budget
Uncertainty budgets for data from the SUV-100 spectroradiometers at SPO and BAR have been published by Bernhard et al. (2004;. Expanded uncertainties (coverage factor k=2, equivalent to uncertainties at the 2σ-level or a confidence interval of 95.45%) for spectral irradiance at 345 nm and erythemal irradi-5 ance vary between 5.8% and 6.2%. These are dominated by uncertainties related to the instruments' radiometric calibration, wavelength error, and cosine-error corrections. A detailed uncertainty budget for the SUV-150B at SUM is available at www.biospherical.com/nsf/Version2/Summit info.asp. The expanded uncertainty is 6.0%, and dominated by radiometric calibration and responsivity drift. The budgets 10 for the two instrument types are very similar due to the large contributions from uncertainties related to the calibration standards (200-Watt tungsten-halogen lamps) used for all spectroradiometers.

Model calculations
Measurements at all sites where complemented with calculations of the radiative trans-15 fer model UVSPEC/libRadtran Version 1.01 (Mayer and Kylling, 2005). The model's pseudospherical disort radiative transfer solver with six streams was used. These calculations are used for quality control, calculation of total ozone, and interpretation of differences seen in measurements at the three sites. Model calculations also provide reference clear-sky spectra. Model input parameters include: solar zenith angle (SZA); 20 the extraterrestrial spectrum; atmospheric profiles of air density, temperature, ozone, and aerosol extinction; total column ozone; effective surface albedo; atmospheric pressure at station level; aerosol optical depth (parameterized by theÅngström coefficients α and β); and single scattering albedo for aerosols (SSA). Settings of these parameters for SPO and BAR are described by Bernhard et al. (2004; Bernhard et al. (2003), and compared with measurements of the Ozone Monitoring Instrument (OMI) on NASA's AURA platform. For SZAs smaller than 80 • , the ratio of SUV-150B and OMI measurements is 1.010±0.019 (±1σ). Surface albedo 5 was set to 0.97 in accordance with measurements by Grenfell and Warren (1994). Station pressure data were provided by NOAA. Aerosol optical depth is not being measured at Summit. Calculations were performed for stratospheric background aerosol conditions by setting α=1.0 and β=0.008. This translates to an aerosol optical depth (AOD) of 0.016 at 500 nm. Actual AODs are likely larger, in particular during spring 10 when Summit may be affected by Arctic haze (Bodhaine and Dutton, 1993). Based on our analysis for BAR (Bernhard et al., 2007), we estimate that aerosols could reduce erythemal irradiance and irradiance at 345 nm on average by 4-5% in spring and 2-3% in summer. SSA was set to 0.99. Figure 2a shows the ratio of irradiance measured during clear skies in the wave-15 length band 342. 5-347.5 nm to the corresponding model result for all sites. Clear-sky periods were determined based on temporal variability as described by Bernhard et al. (2004). Average ratios are 0.959±0.016 (±1σ) for SPO, 0.969±0.030 for BAR, and 0.967±0.024 for SUM. The bias between measurement and model of about 3-4% is within the uncertainty of the measurements and model input parameters. For SPO and 20 BAR, reasons for the discrepancy have been further analyzed by Bernhard et al. (2004;. Data for SUM are also plotted versus time (Fig. 2b) (Manney et al., 2006). It also had a record-size vortex area in early March (Feng et al., 2007). Our analysis of Introduction

Total ozone
Total ozone calculated from measured UV spectra is shown in Fig. 3. There is a stark contrast between the annual cycles of total ozone at SPO and the two northernhemisphere sites. At SPO, the smallest ozone columns are observed in the austral 10 spring (October through early December), when Antarctica is affected by the ozone hole and total ozone is frequently below 220 DU. Total ozone at BAR and SUM is largest between February and April as a result of the Brewer-Dobson circulation (Holton et al., 1995). This phenomenon leads to a poleward transport of ozone from the tropics during the winter and early spring, causing an ozone maximum in spring and a minimum in 15 autumn. However, during years when the stratosphere is cold enough for the formation of extended polar stratospheric clouds (PSC), substantial photochemical destruction of ozone can be observed also in the northern hemisphere (Rex et al., 2004). Such conditions occurred in the winter of 2004/2005 when the PSC area in January 2005 was the largest in 12 years (Feng et al., 2007) Figure 4 shows measurements of spectral irradiance integrated over the range of 342.5-347.5 nm (hereinafter called "irradiance at 345 nm") as a function of SZA for the three sites. Panels a and c are based on all available data; Panels b and d depict the clear-sky subset. The following can be concluded from results presented in Fig. 4: -Measurements at SUM and SPO are very similar ( Fig. 4a and b).

10
-The influence of clouds is very small at SPO and SUM (Fig. 4a) for two reasons: first, low temperatures over the ice caps lead to low atmospheric water content and optically thin clouds. By analyzing SUV-100 measurements in the visible, Bernhard et al. (2004) found that clouds at SPO have an optical depth of smaller than 1 in 71% of the time. This result is also quantitatively supported by data 15 from longwave emission spectroscopy performed at SPO by Mahesh et al. (2001). Second, cloud attenuation is greatly moderated by high albedo due to multiple reflections between the snow-covered surface and clouds (Nichol et al., 2003).
-Measurements at BAR are substantially smaller than at SUM, mostly due to differences in cloudiness and surface albedo (Fig. 4c). The area of the highest 20 point-density in the BAR data set is associated with clear-sky measurements during summer when albedo is low. In summer, clouds at BAR can reduce irradiance at 345 nm by more than 75% (for more details see Bernhard et al., 2007).
-Clear-sky measurements at BAR are clustered into two groups (Fig. 4d)

Printer-friendly Version
Interactive Discussion the albedo is 0.83±0.08. The lower cluster is associated with clear-sky conditions during summer, when albedo is smaller than 5%.
The difference between SPO and SUM was examined in more detail by sorting all clear-sky measurements into 1 • -wide SZA-bins, and forming the ratios of SPO/SUM. Results are shown in Fig. 5. Clear-sky measurements at SPO are larger than SUM 5 data by 7% at SZA=67 • , 5% at 75 • , and 4% at 79 • , on average. This difference, and the small SZA-dependence, can be explained almost entirely by the difference in Earth-Sun distance for the austral and boreal summer (Iqbal et al., 1983). When measurements of both sites are normalized to a Earth-Sun distance of 1 astronomical unit (AU), the ratio of SPO/SUM is 1.015 on average and does not exhibit any significant dependence on SZA (red data set in Fig. 5). This difference is well within the combined uncertainty budget of the two systems, although some contribution may also be due to the tentatively larger atmospheric aerosol loading at SUM.
The differences between BAR and SUM were analyzed in a similar way by binning clear-sky BAR data from the high and low albedo clusters shown in Fig. 4d, and com-15 paring the results with the associated data from SUM. Results are shown in Fig. 6. For the period of high albedo (12-February to 30-April, designated as "spring"), measurements at BAR are on average 14±1% (±1σ) smaller than at SUM. For the period of low albedo (16-June to 30-September, designated as "summer"), measurements at BAR are on average 36±2% below values from SUM, indicating that the difference in 20 albedo between spring and summer contributes about 22% to the difference. (If summer measurements at BAR had been used as the reference, spring measurements at BAR would have been higher by 38%. This is consistent with the analysis presented by Bernhard et al. (2007). Measurements at SUM exceed summer measurements at BAR by about 56%.) 25 The difference between measurements at SUM and BAR during spring are analyzed in more detail as follows. It can be assumed that some of the disparity stems from the differences in altitude (3202 m versus 6 m) and albedo (0.97 versus 0.83±0.08). To quantify the effect, irradiance at 345 nm was modeled for each spectrum measured 4959 Introduction at BAR during clear skies based on the set of model input parameters described in Sect. 5. The results (denoted hereinafter "Run 1") were compared with modified calculations where the model parameters "altitude," "surface pressure," and "albedo" were set to 3202 m, 668 hPa, and 0.97, respectively (Run 2). These parameters represent typical conditions at Summit. The ratio of Run 1/Run 2 is a good estimate of the effect of the three parameters, and is plotted as a red line in Fig. 6. The ratio is approximately 0.89, indicating that measurements at BAR should be 11% smaller than at SUM due to the differences in altitude and albedo. The disparity of about 3% between the modeled and measured ratio (blue line in Fig. 6) could be caused by differences in aerosol loading at the two sites. To investigate this assumption further, a third model run (Run 3) was executed by setting model aerosol parameters to background aerosol conditions (α=1.0 and β=0.008), and otherwise using the same modified parameters for altitude, pressure and albedo that were implemented for Run 2. Note that results shown in Fig. 2b have suggested that SUM is also affected by aerosols, in particular during winter and early spring. Model calculations with background conditions are therefore not 15 meant to be a simulation for the actual aerosol loading at SUM (which is not known from measurements), but are instead an estimate of the maximum influence that can be expected from aerosol attenuation. The ratio of Run 1/Run 3 is plotted as a green line in Fig. 6. The ratio is 4-6% below the ratio of Run 1/Run 2 discussed earlier, confirming results by Bernhard et al. (2007) 20 that aerosols at BAR reduce irradiance at 345 nm by several percent. The measured ratio (blue line) is between the ratios of the model runs. This indicates that the actual aerosol attenuation at SUM is likely between the reductions expected for background conditions and the condition prevailing at BAR. Figure 7 shows measurements of the UV Index as a function of SZA for the three sites. The UV Index was calculated by weighting measured spectra with the CIE action spectrum for erythema (sunburn) (McKinlay and Diffey, 1987)  were again partitioned into all-sky ( Fig. 7a and c) and clear-sky (Fig. 7b and d) cases.

25
The following can be concluded: -UV Indices are mostly controlled by the SZA.
-The overall maximum UV Index is 6.7 at SUM, 5.0 at BAR and 4.0 at SPO. At 5 SZA=70 • , UV Indices vary between 0.8 and 1.9 at SUM, 0.0-1.2 at BAR, and 0.9 and 3.4 at SPO.
-For comparable SZAs, UV Indices measured at SPO during the period of the ozone hole are approximately twice as large as maximum indices observed at SUM ( Fig. 7a and b). For times not affected by the ozone hole, measurements at 10 SPO are comparable to maximum indices at SUM, but the majority of measurements at SUM are considerably below SPO levels.
-UV Indices at SUM exceed UV Indices at BAR by more than 50% on average ( Fig. 7c and d).
-At SUM, the annual cycle in total ozone (Fig. 3) leads to a distinct seasonal UV 15 Index variability. The lower halves of the ranges shown in Fig. 7c and d are mostly composed of measurements performed between February and May, whereas the upper halves are based on data collected between June and October (see also Fig. 8).
-Clear-sky UV Indices at BAR (Fig. 7d) are not clustered into two groups as was 20 the case for irradiance at 345 nm (Fig. 4d). This is a consequence of the annual albedo and ozone cycles: enhancement of irradiance by larger albedo during February-April is balanced by the increased absorption by ozone during these months, leading to similar clear-sky UV Indices year-round for a given SZA. The relative importance of these two factors has been quantified in more detail by The differences in UV Indices at the three sites were analyzed in more detail by dividing the measurements shown in Fig. 7c into spring and summer periods, and sorting the resulting datasets into to 1 • -wide SZA-bins. Results are shown in Fig. 8a. For SZA>68 • , the largest UV Indices are observed at SPO during spring . There is little variability (indicated by errors bars) for SZAs larger than 5 80 • (19-October at SPO), because SPO is affected by the ozone hole nearly during all days of October. Considerably more variability is observed at smaller SZAs (November) due to large year-to-year changes in total ozone for this month (Fig. 3). UV Indices observed at SPO during summer (1-January-30-March) again show little variability.
UV Indices at SUM are generally smaller than at SPO for comparable SZAs. Mea- measurements between BAR and SUM is mostly due to the difference in albedo and altitude discussed earlier. Summer measurements at BAR are reduced due to small albedo and considerable cloud influence, leading to substantial variability in the UV Index as indicated by the large error bars in Fig. 8a.
To emphasize relative differences between the three sites at large SZAs, all data 20 sets shown in Figure 8a were ratioed against summer measurements at SUM, with results presented in Figure 8b. The most striking feature is the large ratio for spring-time measurements at SPO. The ratio peaks at SZA=84 • (8-October at SPO), when measurements at SPO are factors of 2.3 and 5.0 larger than at SUM and BAR, respectively. The peak is a consequence of the small ozone column observed at SPO on this day Ratios for the other data sets shown in Fig. 8b are 1.2-1.3 for "SPO summer"; 0.8-1.0 for "SUM spring"; 0.6-0.8 for "BAR spring"; and 0.4-0.5 for "BAR summer". Ratios for spring generally have a larger slope than those for summer as days with photochemically-induced spring-time ozone depletion typically occur early in the year and contribute mostly to bins with large SZA. One example of this effect is the period 5 15-February-15-March of 2005 when total ozone at SUM was below 350 DU. UV Indices from this period were binned and ratioed against summer measurements like the other data sets. The resulting ratios are indicated by a broken line in Fig. 8b. The period only contributes to bins with SZAs larger than 74 • , and ratios are 20% on average larger than those of the "SUM spring" data set.

Conclusions
Measurements from the new UVSIMN network site at SUM were compared with similar measurements at SPO and BAR. Data were used for investigating the factors influencing solar UV radiation at high latitudes, and for assessing differences in the UV climate between the northern and southern hemisphere. When corrected for the difference in 15 Earth-Sun distance, measurements of irradiance at 345 nm were almost identical at SUM and SPO. The influence of clouds is very small at the two sites due to low atmospheric water content and high surface albedo. Clear-sky measurements at BAR are substantially smaller than at SUM (i.e., 14% in spring and 36% in summer), mostly due to differences in surface albedo, altitude, and aerosols. Comparisons with model 20 calculations have indicated that aerosols can reduce clear-sky irradiance at 345 nm by 4-6%. The aerosol influence is largest in April and tapers off as the year progresses. This is consistent with the annual cycle of Arctic haze reported by Bodhaine and Dutton (1993 SPO, ozone columns are substantially reduced during the ozone hole period between September and November. UV Indices measured during these months are increased by 20-80% on average compared to measurements between January and March. The environment at SUM is quite the opposite. Total ozone peaks in April and UV Indices in spring are on average 10-25% smaller than in the summer. One exception is the 5 unusually cold winter of 2004/2005 (Feng et al., 2005), which led to large chemical ozone losses in February and March of 2005. This caused large relative enhancements of the UV Index at SUM. Absolute gains remained below 0.1 UV Index unit however, due to the low solar elevations prevailing during this period, which is inconsequential for all practical purposes. 10 The maximum UV Index historically observed at SUM was 6.7, which exceeds the maximum UV Index at SPO by more than 50%. This is mostly a consequence of SUM's lower latitude and larger solar elevations. For comparable SZAs, average UV Indices measured during October and November at SPO are 1.9-2.4 times larger than measurements during March and April at SUM. UV Indices at SUM exceed those at 15 BAR by over 50%, again mostly due to larger cloud influence at BAR.
The conditions (e.g., albedo, altitude, aerosol influence, clouds) observed at SUM and SPO are representative for the Greenland ice cap and Antarctic plateau. We believe that results presented here can be generalized to other locations on the polar plains. Extrapolation to coastal places is more problematic due to the influence of lo-20 cal effects, notably topography, inhomogeneous albedo, pollution, fog, and clouds as observed at BAR. Additional monitoring sites at high latitudes, particularly in Russia, would be required for a more complete assessment of the Arctic UV climate. We anticipate that continuance of measurements at SUM will prove helpful in assessing future changes of the Arctic UV levels, in particular with respect to changes in stratospheric 25 temperatures, ozone (column and profile), aerosols, and atmospheric circulation patterns.