Reconstruction of Solar Spectral Surface UV Irradiances UsingRadiative Transfer Simulations

Anders Lindfors*1, Anu Heikkila1, Jussi Kaurola1, Tapani Koskela1 and Kaisa Lakkala2

1Climate Change Research, Finnish Meteorological Institute, Helsinki, Finland2Arctic Research Centre, Finnish Meteorological Institute, Sodankyla, Finland

Received 5 November 2008, accepted 2 April 2009, DOI: 10.1111 ⁄ j.1751-1097.2009.00578.x

ABSTRACT

UV radiation exerts several effects concerning life on Earth, and

spectral information on the prevailing UV radiation conditions is

needed in order to study each of these effects. In this paper, we

present a method for reconstruction of solar spectral UV

irradiances at the Earth’s surface. The method, which is a

further development of an earlier published method for recon-

struction of erythemally weighted UV, relies on radiative

transfer simulations, and takes as input (1) the effective cloud

optical depth as inferred from pyranometer measurements of

global radiation (300–3000 nm); (2) the total ozone column; (3)

the surface albedo as estimated from measurements of snow

depth; (4) the total water vapor column; and (5) the altitude of

the location. Reconstructed daily cumulative spectral irradiances

at Jokioinen and Sodankyla in Finland are, in general, in good

agreement with measurements. The mean percentage difference,

for instance, is mostly within ±8%, and the root mean square of

the percentage difference is around 10% or below for wave-

lengths over 310 nm and daily minimum solar zenith angles

(SZA) less than 70�. In this study, we used pseudospherical

radiative transfer simulations, which were shown to improve the

performance of our method under large SZA (low Sun).

INTRODUCTION

Recent assessment reports by the United Nations Environ-mental Panel (1–3) have clearly brought forth that solar UVradiation exerts important effects concerning life on Earth; UV

radiation affects not only human health, but also terrestrialand aquatic ecosystems, materials and atmospheric chemistry.In order to study these effects in detail, spectral information on

the ambient radiation is needed.Because of the demanding nature of performing and

maintaining spectral surface UV measurements, they have alimited temporal and spatial coverage. On the other hand,

pyranometer measurements of global radiation (300–3000 nm)are more commonly available, and it has been shown (e.g. 4–7)that pyranometer-based UV reconstruction techniques provide

accurate information on the surface UV radiation. So far,however, these reconstruction efforts have concentrated onselected action spectra (mostly erythemally weighted) or

wavelengths.

Interestingly, Lindfors et al. (7) noted that their UVreconstruction method could also be further developed toproduce the whole UV spectrum. In the present paper, we

present this spectral UV reconstruction method, and validatethe results using measurements by the Brewer spectrophotom-eters at Jokioinen and Sodankyla in Finland.

METHOD AND INPUT DATA

The spectral UV reconstruction method presented in this paper is afurther development of the method for reconstruction of erythemallyweighted UV by Lindfors et al. ([7]; we will hereinafter refer to theirpaper as L07). In L07, the idea was to use pyranometer measurementsfor determining the influence of clouds on UV radiation. Based onphysical relationships determined through radiative transfer modeling,they developed a look-up-table enabling transfer of the cloudinformation contained in the pyranometer measurements into a cloudeffect in the UV range.

As discussed in L07, the cloud optical depth can also be explicitlyinferred from measurements of global radiation (see also, e.g. [8,9]).This cloud optical depth can then be given as input for radiativetransfer simulations aiming to produce the whole UV spectrum,similarly to that done by Dahlback (10) based on multichannel filterradiometer measurements.

Spectral UV reconstruction. We simulated, or reconstructed, thespectral surface UV irradiance for two stations in Finland, namelyJokioinen (60.8�N, 23.5�E, 104 m a.s.l.) and Sodankyla (67.4�N,26.6�E, 179 m a.s.l.), using the libRadtran radiative transfer package(11). In our radiation transfer simulations, we assumed a horizontallyhomogeneous water cloud layer extending from 2 to 4 km in altitude,with an effective cloud droplet radius of 10 lm (as in L07). The cloudoptical depth, which was inferred from the measured global radiationas explained more in detail beneath, was set at the wavelength 550 nm,and its wavelength dependence was determined through the parame-terization by Hu and Stamnes (12). Other input to our radiativetransfer simulations were as described in L07 (see their paper for moredetails): daily total ozone column as measured by the Brewerinstrument; daily effective surface UV albedo as estimated from snowdepth measurements according to a satellite-based method (13); andrural background aerosol properties, with an aerosol optical depth setto follow a climatological annual cycle showing low values (0.10 at310 nm) in the winter and higher values (0.26) in the summer. It can bementioned further that the midwinter UV albedo typically took valuesaround 0.6 and 0.4 at Sodankyla and Jokioinen, respectively, while thesummer UV albedo was set to 0.05. Finally, we assumed midlatitudestandard atmospheric constituent profiles ([14]; we used their summerprofile for April–September, and winter profile for October–March)and convoluted the simulated UV spectra with the slit function of therespective Brewer instruments at Sodankyla and Jokioinen in order tomake them comparable to measured spectra. Note that at Jokioinen,the optics of the monochromator of the Brewer were realigned between24 and 25 April 1997 (15). Therefore, we used two different slitfunctions for Jokioinen, one for the period up to 24 April 1997, andone for the period thereafter.

*Corresponding author email: anders.lindfors@fmi.fi (Anders Lindfors)� 2009TheAuthors. JournalCompilation.TheAmericanSociety ofPhotobiology 0031-8655/09

Photochemistry and Photobiology, 2009, 85: 1233–1239

1233

L07 noted in their work a systematic underestimation of the UVirradiances under large solar zenith angles (SZA). Furthermore, theyfound that a major part of this underestimation can be attributed tothe plane-parallel approximation used in the radiative transfersimulations, which does not properly account for the spherical shapeof the atmosphere. In order to improve this, we here used thepseudospherical sdisort solver of the radiative transfer equation(included in libRadtran), with the number of streams set to 10.

Cloud optical depth from global radiation. For determining the cloudoptical depth at 550 nm, we used hourly values of global radiationfrom Kipp and Zonen CM11 pyranometer measurements (see L07 formore details on the data). Following the idea of earlier work (8,9), wedeveloped a look-up-table for retrieving the cloud optical depth. Thelook-up-table was produced by radiative transfer simulations of thesurface global radiation with varying cloud optical depth (s550), SZA,surface albedo (a) and total water vapor column (TWC; because watervapor absorbs in the infrared part of the global radiation). The globalradiation was simulated using an optimized version (kato2; included inlibRadtran) of the correlated-k method (16). Otherwise, the setup ofthe radiative transfer simulations was as described above. This meansthat the look-up-table for retrieving the cloud optical depth at 550 nmcan be expressed as the function f in

s550 ¼ fðGmeas; SZA; a;TWC)

where Gmeas is the measured hourly global radiation corrected to thesame Earth–Sun distance as used when the look-up-table was pro-duced. It can be mentioned further that a was set to 0.12 for snow-freeconditions, and assumed equal to the UV albedo for snow-coveredconditions (following L07), while TWC was taken from the ERA-40reanalysis or the ECMWF model analysis (also following L07).

It is emphasized that the applied radiative transfer model is one-dimensional and as such it represents a simplification of the realatmosphere. For instance, it does not allow for three-dimensionalcloud structures. Therefore, the retrieved cloud optical depth repre-sents an effective cloud optical depth (see also [9]). This effective cloudoptical depth, however, provides useful and accurate information onthe effect of clouds on radiation, and can hence be used forreconstruction of spectral UV irradiances.

VALIDATION

Using the method described above, we reconstructed hourlyspectral surface UV irradiances for Jokioinen and Sodankyla

for the periods 1995–2006 and 1990–2006, respectively, as forthese periods measurements of the surface UV irradiance wereavailable. In this section, we present validation results achieved

through a comparison against measured UV irradiances.

Reference data

Spectral UV irradiances are being measured with well-main-

tained Brewer spectrophotometers at Jokioinen andSodankyla. At Sodankyla, a single monochromator Brewer isdeployed, covering the wavelength range 290–325 nm, whilethe double monochromator Brewer at Jokioinen measures

radiation up to 365 nm. These measurements were recentlydescribed in more detail by Lakkala et al. (15). The Brewer atJokioinen meets the WMO level S2 requirements for detection

of trends in the surface UV irradiance (17). Although no totaluncertainty budget is available for the Brewer instruments atJokioinen and Sodankyla, an earlier study (18) considered the

daily erythemal UV doses inferred from these data to beaccurate within at least 10%.

For comparing reconstructed and measured UV, we used

daily cumulative spectral irradiances. According to Heikkilaet al. (19), this quantity could be suitable for studies on theeffects of accumulated UV radiation exposure, for instance,

UV radiation-induced degradation of materials in outdoorweathering experiments. The advantage of this quantity is thatit retains the spectral information, which means that it can bemultiplied by an action spectrum derived for any type of

(damaging) effect. From a validation perspective, it is alsomore meaningful to use daily cumulative spectral irradiances,because a direct comparison between measured momentary

spectra and reconstructed hourly spectra is not straightfor-ward due to differences in temporal representativeness (asexplained more in detail in L07). However, we emphasize that

hourly reconstructed spectra are available and can be used forstudies that need data of higher temporal resolution. Finally, itis worth mentioning that some validation for hourly

erythemally weighted UV was presented in L07.The Brewer spectrophotometers at Jokioinen and

Sodankyla have a typical scan frequency of once or twice anhour, thus generally covering the day fairly well. Occasionally,

however, there may be disturbances in the scan schedule, forinstance, due to instrument maintenance. In order to excludedays not well covered by the Brewer measurements from our

analysis, we applied the following selection criteria whencalculating the daily cumulative irradiances from the measureddata: (1) we required that the time gap (SZA gap) between

consecutive scans should not exceed 2 h (20�) and (2) that thesolar elevation angle during the first and last scans of the daymust not exceed 30% of the maximum solar elevation duringthe day. Finally, we assumed a sensitivity of 10)4 and

10)6 W m)2 nm)1, respectively, for the single (Sodankyla)and the double (Jokioinen) monochromator Brewer. Irradi-ance values below these limits were not included when

calculating the daily accumulation, because they were consid-ered not to be distinguished from the noise. In practice, theseselection criteria meant for Sodankyla that the dark winter

period between early November and early February (with dailyminimum SZA > �83�) was not included in our analysis.Thus, potentially viable ozone measurements performed under

these conditions of very low Sun and used as input to ourradiative transfer simulations were also excluded.

Figure 1 shows, as an example, single measured UVirradiance spectra in the morning and at noon, together with

the daily cumulative spectrum for a selected winter andsummer day at Jokioinen. The midday values contribute amajor part to the daily cumulative irradiance, in particular at

the shorter wavelengths. Comparing the noon spectrum withthe morning spectrum highlights the combined effect of ozoneabsorption and SZA; with higher Sun (noon), the short

wavelengths (under 320 nm) are much more pronounced. Itcan be mentioned further, that in winter (December–February), an average of seven scans per day contributed tothe daily cumulative UV spectrum, while in summer (June–

August), the number of single spectra contributing to the dailyaccumulation was 25 on average (these numbers include datafrom both Jokioinen and Sodankyla).

Statistics

For assessing the performance of our method, we used the rootmean square (RMS) of the percentage difference as a measure

of the spread between the reconstructed and the measuredcumulative spectral irradiances, and the mean percentagedifference for the systematic deviation. The definition of both

1234 Anders Lindfors et al.

the RMS and the mean difference can be found in L07 (their

eqs. 5 and 6). Finally, we also included the correlationcoefficient as a general measure of the agreement.

RESULTS

As an example illustrating the general performance of themethod, Fig. 2 shows a time series of reconstructed and

measured daily cumulative irradiances at the wavelength300 nm (Fig. 2a) and 320 nm (Fig. 2b) during 2001 atJokioinen. At both wavelengths, the reconstructed irradiances

follow closely the measured ones, although at 300 nm, wherethe irradiance is roughly two decades lower than at 320 nm,somewhat larger deviations from the measured values can be

seen.The scatter plots in Fig. 3, where the measured and

reconstructed daily cumulative irradiances at the wavelength310 nm are compared for both Jokioinen (Fig. 3a) and

Sodankyla (Fig. 3b), confirm the general good performance

of the reconstruction method. All points are confined to arelatively narrow region concentrating around the diagonalaxis and the systematic deviation is small at both stations,being within a few percent. At Jokioinen, however, a weak

tendency toward overestimation at moderate cumulativeirradiances (around 1000 J m)2 nm)1) can be noted. The smallsystematic difference between the reconstructed and measured

UV irradiances could, in principle, be explained by, forinstance, aerosol or albedo input to the radiative transfersimulations that may not correspond perfectly to reality.

However, taking into account the uncertainty of the measure-ments themselves (�10%), we find the agreement satisfactory.The RMS difference between the reconstructed and measured

daily cumulative irradiances is 11% and 13%, respectively, forJokioinen and Sodankyla, which is slightly higher than thatfound for the erythemally weighted daily doses in L07, wherethe RMS was mostly below 10%. Note that Fig. 3, as well as

Figs. 4–6, contain data for the whole period included in thisstudy, that is, 1995–2006 at Jokioinen and 1990–2006 atSodankyla.

Figure 4 shows for each wavelength the number of days forwhich the measurements yield a nonzero cumulative irradiance(according to the criteria described above). A transition from a

low to a high number of nonzero daily cumulative irradiancestakes place around 300 nm for the single monochromatorBrewer at Sodankyla, and around 295 nm for the doublemonochromator Brewer at Jokioinen. Thus, we limit the

comparison of the reconstruction-based and measurement-based daily spectral cumulative UV irradiances to the wave-length range 300–325 and 300–365 nm for Sodankyla and

Jokioinen, respectively (the same lower limit was used also in[20]). Note that the lower amount of days with nonzerocumulative irradiance for wavelengths above 363 nm at

Jokioinen is due to the realignment of the monochromatormentioned above.

Figure 5, in turn, shows the correlation coefficient between

the reconstructed and the measured cumulative daily irradi-ances as a function of wavelength. In general, the correlation isvery satisfactory, being above 0.99 throughout the consideredwavelength range with only occasional exceptions (e.g. at some

wavelengths near 300 nm at Jokioinen). The correlationcoefficient peaks around 310 nm at both stations, decreasingslowly toward longer wavelengths and more rapidly toward

shorter wavelengths. In particular at Jokioinen, some individ-ual wavelengths (e.g. 319 and 350 nm) show a lower correla-tion than their neighbors. This is likely connected to the very

fine structure in the spectral data that are compared (i.e.reconstructed versus measured spectral data), which is dis-cussed in more detail later. It is emphasized that the analyzeddata span a broad range of different SZA, which strongly

drives the irradiance levels and thus is partly responsible forthe high correlation coefficients. Still, the correlation coeffi-cient does provide a rough measure of the performance of the

method.Figure 6 shows for both stations the RMS difference and

the mean difference of the reconstructed daily cumulative

spectral irradiances compared to measurements. The data havebeen divided into subsets according to the minimum SZA ofthe day (SZAmin). For SZAmin < 70� and wavelengths

>310 nm, the RMS difference is around 10% or somewhatless at both stations. The mean difference is mostly within

Figure 1. Measured daily cumulative irradiance at Jokioinen togetherwith morning and noon spectra for a (a) winter and a (b) summer day,respectively. The sky was nearly cloud free during both days. On 16February 2005 the total ozone column was 311 DU and the Brewerscans shown in the figure were taken at a solar zenith angle of 85.8�(�6:50 UTC) and 73.0� (�10:40 UTC). On 16 July 2005 the totalozone column was 312 DU and the scans were taken at 86.0� (�2:20UTC) and 39.5� (�10:30 UTC).

Photochemistry and Photobiology, 2009, 85 1235

±8% and shows only a small systematic dependence onSZAmin, indicating that the pseudospherical radiative transfer

simulations used in this work improve the underestimationproblem seen for large SZA in L07.

At wavelengths lower than 310 nm, the RMS difference

(Fig. 6a,b) increases with decreasing wavelength in allSZAmin groups. At these wavelengths, the irradiance ismuch lower than at longer wavelengths (see Fig. 2), which

makes it more difficult to accurately simulate the irradiancelevel as the relative difference may easily become large. Theuncertainty of the measured data also increases withdecreasing signal strength. This is mainly due to photon

counting uncertainty and directional errors at large SZA.Furthermore, photons at the short wavelengths are stronglyabsorbed by ozone, meaning that, for instance, ozone

variations that occur within a day may also contribute tothe higher RMS (as the method uses daily ozone values asinput).

The mean difference (Fig. 6c,d) between the reconstructedand the measured daily cumulative irradiances fluctuatesrapidly from one wavelength to another, the general pattern

being very similar for all SZAmin groups. Moreover, somefeatures, for instance the peak at 309 nm, are present both atSodankyla and Jokioinen. Thus, it is clear that this rapidfluctuation is systematic. Also other studies (20–22) have

found a similar fine structure when comparing simulated andmeasured spectra. The reason for this behavior is likelyconnected to the very fine structure in the extraterrestrial solar

spectrum caused by the Fraunhofer absorption lines incombination with the limitations (e.g. exact knowledge of the

slit function and wavelength scale) set by both the ground-based instrumentation and by the space-born instrumentationmeasuring the extraterrestrial irradiance used as input to our

radiative transfer simulations (see also [20,21]). It can bementioned further, that this spectral fine structure was clearlyless pronounced in an earlier study (23) where the recon-

structed irradiances were compared against measurementsthat had been corrected for wavelength shifts and stan-dardized to a 1 nm slit function using the SHICrivm tool(http://www.rivm.nl/shic).

At Sodankyla (Fig. 6d), there is a systematic behaviordepending on wavelength as the mean difference betweenreconstructed and measured cumulative spectral irradiances

decreases with decreasing wavelength below 310 nm for thegroups SZAmin < 70�. Part of this can probably be attributedto the typical single monochromator stray light problem

causing the measured irradiances to be too high at shortwavelengths. A similar behavior has been noted for theSodankyla Brewer also in comparison with reference

instruments (15).

DISCUSSION

In this paper, we present a method for reconstruction of solarspectral UV irradiances using pyranometer measurements ofglobal radiation. The method relies on radiative transfer

Figure 2. Reconstructed and measured daily cumulative irradiances at (a) 300 nm and (b) 320 nm at Jokioinen during 2001. Note that themeasured cumulative irradiance is missing for the darkest winter period (roughly November–January) because of the selection criteria described inthe text.

1236 Anders Lindfors et al.

simulations and is a further development of the method forreconstruction of erythemally weighted UV presented in L07.

As the methods are very similar, we refer the interested readerto L07 for a more detailed discussion on the physical aspects ofthe method.

Reconstructed daily cumulative spectral irradiances werevalidated against Brewer measurements at Jokioinen andSodankyla in Finland. The results show, in general, a goodagreement with measurements. The mean percentage differ-

ence, for instance, is mostly within ±8%. This is of the sameorder of magnitude as the expected uncertainty of themeasurements. The RMS of the percentage difference, in turn,

is around 10% or below for wavelengths over 310 nm andminimum SZA of the day less than 70�. As the absoluteirradiances decrease rapidly toward shorter wavelengths and

higher SZA, a good agreement on the relative scale becomesmore difficult to achieve. Therefore, the RMS difference ishigher under these conditions.

As an improvement compared to L07, we here used

pseudospherical radiative transfer simulations, which areexpected to give better agreement with measurements under

conditions of large SZA (low Sun). Indeed, our validationresults show only a small systematic difference for allminimum SZA groups (Fig. 6c,d).

Our UV reconstruction method is general in the sense that it

can be applied to a broad range of different locations.However, it would still be important to continue validatingthe method also at other stations, in particular in other

climatic regions. The method aims to serve studies on theeffects of UV radiation by providing the possibility to producereconstructed spectral UV irradiances or dose rates using a

desired action spectrum for any location where pyranometermeasurements are available. As the method provides spectralirradiances up to 400 nm, or even further if wanted, it is

straightforward to apply different action spectra to thereconstructed data. Already, the method has been utilized ina study on the UV effects on alder and birch leaf phenolics(24).

Figure 3. Measured versus reconstructed daily cumulative irradiancesat 310 nm for (a) Jokioinen and (b) Sodankyla. Also shown are thecorrelation coefficient, the RMS difference, the mean difference andthe number of days (n) included in the analysis.

Figure 4. Number of days with nonzero measured daily cumulativeirradiance at Jokioinen and Sodankyla, respectively, as a function ofwavelength.

Figure 5. The correlation coefficient between measured and recon-structed daily cumulative irradiances at Jokioinen and Sodankyla,respectively, as a function of wavelength.

Photochemistry and Photobiology, 2009, 85 1237

Acknowledgements—This work was supported by the European

Commission-funded SCOUT-O3 project (505390-GOCE-CT-2004)

and by the Finnish Funding Agency for Technology and Innovation,

Oy All-Plast Ab, Exel Oyj, MacGregor (FIN) Oy and Nokian Tyres

plc through the UVEMA project. We thank Esa Saarinen, Kimmo

Rikkonen and Jouni Jutila at Jokioinen as well as the operators at

Sodankyla for their efforts to maintain the Brewer spectroradiometer

measurements. We also thank the COST Action 726 for providing an

international forum for work related to UV reconstruction.

REFERENCES1. UNEP (1998) Environmental effects of ozone depletion: 1998

assessment. J. Photochem. Photobiol. B, Biol. 46, 1–108.2. UNEP (2003) Environmental effects of ozone depletion and its

interactions with climate change: 2002 assessment. Photochem.Photobiol. Sci. 2, 1–72.

3. UNEP (2007) Environmental effects of ozone depletion and itsinteractions with climate change: 2006 assessment. Photochem.Photobiol. Sci. 6, 201–332.

4. Fioletov, V., L. McArthur, J. Kerr and D. Wardle (2001) Long-term variations of UV-B irradiance over Canada estimated fromBrewer observations and derived from ozone and pyranometermeasurements. J. Geophys. Res. 106, 23009–23027.

5. Feister, U., E. Jakel and K. Gericke (2002) Parameterization ofdaily solar global ultraviolet irradiation. Photochem. Photobiol. 76,281–293.

6. den Outer, P. N., H. Slaper and R. B. Tax (2005) UV radiation inthe Netherlands: Assessing long-term variability and trends inrelation to ozone and clouds. J. Geophys. Res. 110, D02203.DOI: 10.1029/2004JD004824.

7. Lindfors, A., J. Kaurola, A. Arola, T. Koskela, K. Lakkala, W.Josefsson, J. A. Olseth and B. Johnsen (2007) A method forreconstruction of past UV radiation based on radiative transfermodeling: Applied to four stations in northern Europe. J. Geo-phys. Res. 112, D23201. DOI: 10.1029/2007JD008454.

8. Leontieva, E., K. Stamnes and J. Olseth (1994) Cloud opticalproperties at Bergen (Norway) based on the analysis of long-termsolar irradiance records. Theor. Appl. Climatol. 50, 73–82.

9. Barker, H. W., T. J. Curtis, E. Leontieva and K. Stamnes (1998)Optical depth of overcast cloud across Canada: Estimates basedon surface pyranometer and satellite measurements. J. Clim. 11,2980–2994.

10. Dahlback, A. (1996) Measurements of biologically effective UVdoses, total ozone abundances, and cloud effects withmultichannel, moderate bandwidth filter instruments. Appl. Opt.35, 6514–6521.

11. Mayer, B. and A. Kylling (2005) Technical note: The libRadtransoftware package for radiative transfer calculations—Descriptionand examples of use. Atmos. Chem. Phys. 5, 1855–1877.

Figure 6. The RMS difference and mean difference between reconstructed and measured daily cumulative irradiances at Jokioinen and Sodankyla,respectively, as a function of wavelength. The data were divided into four groups depending on the minimum solar zenith angle of the day(SZAmin).

1238 Anders Lindfors et al.

12. Hu, Y. and K. Stamnes (1993) An accurate parameterization ofthe radiative properties of water clouds suitable for use in climatemodels. J. Clim. 6, 728–742.

13. Arola, A., J. Kaurola, L. Koskinen, A. Tanskanen, T. Tikkanen,P. Taalas, J. Herman, N. Krotkov and V. Fioletov (2003) A newapproach to estimating the albedo for snow-covered surfaces inthe satellite UV method. J. Geophys. Res. 108, 4531. DOI:10.1029/2003JD003492.

14. Anderson, G. P., S. Clough, F. Keizys, J. Chetwynd and E. Shettle(1986) AFGL atmospheric constituent profiles (0–120km), no.954. In Environmental Research Papers, pp. 1–43. Air ForceGeophysics Lab, Hanscom Air Force Base, Bedford, MA.

15. Lakkala, K., A. Arola, A. Heikkila, J. Kaurola, T. Koskela,E. Kyro, A. Lindfors, O. Meinander, A. Tanskanen, J. Grobnerand G. Hulsen (2008) Quality assurance of the Brewer UVmeasurements in Finland. Atmos. Chem. Phys. 8, 3369–3383.

16. Kato, S., T. P. Ackerman, J. H. Mather and E. E. Clothiaux(1999) The k-distribution method and correlated-k approximationfor a shortwave radiative transfer model. J. Quant. Spectrosc.Radiat. Transfer 62, 109–121.

17. Seckmeyer, G., A. Bais, G. Bernhard, M. Blumthaler, C. R.Booth, P. Disterhoft, P. Erikson, R. L. McKenzie, M. Miyauchiand C. Roy (2001) Instruments to Measure Solar UltravioletRadiation, Part 1: Spectral Instruments, Rep. 125, WMO TD 1066.World Meteorological Organization, Geneva, Switzerland.

18. Tanskanen, A., A. Lindfors, A. Maatta, N. Krotkov, J. Herman,J. Kaurola, T. Koskela, K. Lakkala, V. Fioletov, G. Bernhard,

R. McKenzie, Y. Kondo, M. O’Neill, H. Slaper, P. den Outer,A. Bais and J. Tamminen (2007) Validation of daily erythemaldoses from ozone monitoring instrument with ground-based UVmeasurement data. J. Geophys. Res. 112, D24S44. DOI: 10.1029/2007JD008830.

19. Heikkila, A., A. Tanskanen, P. Karha and K. Hanhi (2007)Adjusting timing of weathering test to account for seasonal vari-ations in UV exposure. Polym. Degrad. Stab. 92, 675–683.

20. Bernhard, G., C. R. Booth and J. C. Ehramjian (2004) Version 2data of the National Science Foundation’s Ultraviolet RadiationMonitoring Network: South Pole. J. Geophys. Res. 109, D21207.DOI: 10.1029/2004JD004937.

21. Mayer, B., G. Seckmeyer and A. Kylling (1997) Systematic long-term comparison of spectral UV measurements and UVSPECmodeling results. J. Geophys. Res. 102, 8755–8767.

22. Wuttke, S., J. Verdebout and G. Seckmeyer (2003) An improvedalgorithm for satellite-derived UV radiation. Photochem.Photobiol. 77, 52–57.

23. Koskela, T., A. Heikkila, J. Kaurola, A. Lindfors, A. Tanskanenand P. den Outer (2006) Spectral UV monitoring—Worth it?In Remote Sensing of Clouds and the Atmosphere XI. (Edited byJ. R. Slusser, K. Schafer and A. Comeron), Proc. SPIE Int. Soc.Opt. Eng. 6362, 63622A.

24. Kotilainen, T., R. Tegelberg, R. Julkunen-Tiitto, A. Lindfors andP. Aphalo (2008) Metabolite specific effects of solar UV-A andUV-B on alder and birch leaf phenolics. Glob. Chang. Biol. 14,1294–1304.

Photochemistry and Photobiology, 2009, 85 1239