Norwegian University of Science and Technology, Department of Physics, N-7034 Trondheim, Norway.
* To whom correspondence should be addressed. (email: Trond.Thorseth@phys.ntnu.no)

Accurate spectral solar irradiance measurements depend upon a good calibration. If the spectroradiometer operates at a different temperature than the calibration temperature, sensitivity and wavelength setting of the instrument will change. This is the case for most spectroradiometers, but the magnitude of this effect will vary for different types of instruments. In this study a widely used spectroradiometer, Optronic OL752,  from Optronic Laboratories Inc. has been investigated. Both the effect of temperature on sensitivity and wavelength shift has been quantified. The sensitivity of the Optronic OL752 in the ultraviolet region, 290-400 nm, decreased approximately 0.8 %/K when the instrument was heated above the calibration temperature. Wavelength shift coefficients were derived for several wavelength regions with two independent methods. One of the methods used was direct observation of how the measured spectral lines from a Hg lamp, shifted with temperature. The other method, calculated the relative shift in the spectral global irradiance measurements by comparing the Fraunhofer structure in the measured spectra to the structure in an extraterrestrial spectrum. OL752 had a systematic wavelength shift, varying from 0.10 nm/K  at 254 nm  to 0.06 nm/K at 633 nm. CIE-weighted irradiance derived from modeled data, was used to estimate errors due to temperature effects. If the spectroradiometer temperature is not controlled, a change of 5 K may lead to an error in the derived CIE-weighted irradiance from 10% to 15% dependent upon solar zenith angle and atmospheric conditions.

INTRODUCTION

The knowledge about decreased global total atmospheric ozone (1)(2)(3) and the consequences of increased UV radiation (4)(5) have caused a need to measure solar UV radiation accurately. Many countries have establish UV networks,  monitoring solar UV irradiation continuously, either with spectroradiometers, broadband meters or multichannel instruments (6). There has been an increased number of field experiments in which the effects of  increased UV have been investigated on different  biological systems (7) and there is a need to establish UV-dose rates. Accurate measurements of UV are complicated and the overall uncertainty can be due to several factors (8).  One of these factors is the ambient temperature. It is well known that the instrument response changes with temperature, both spectroradiometers (9)(10)(11) , broadband instruments (12) and narrow channel instruments (13). To avoid the effect of fluctuating  temperatures,  many UV monitoring instruments are temperature stabilized (14) or the effect is taken into account in the measuring procedure in different ways (15)(16). If the effects of temperature changes are not taken into account or the temperature control fails, temperature dependent error will be introduced in the UV irradiance measurements.

For most spectroradiometers both the wavelength calibration and the sensitivity calibration are temperature dependent. The calibration should be performed at the same temperature as the measurements are done. For non-temperature controlled spectroradiometers this is not always possible. During field experiments, the temperature might change several degrees, and the uncertainty in the UV irradiance will increase accordingly. The aim of the present paper is to quantify the error introduced  in solar spectral UV measurements, when the spectroradiometer temperature fluctuates.  This is done for a frequently used spectroradiometer, Optronic OL752. Wavelength dependent temperature coefficients will be derived for wavelength shift and sensitivity.

Biological dose rates can be derived from spectral solar irradiance spectra, when the biological action spectrum is known. The uncertainty in biological dose rates will depend both on the irradiance measurements and the action spectrum. The CIE reference action spectrum  for ultraviolet induced erythema in human skin, is a standardized action spectrum proposed by the International Commission on Illumination (CIE) (17).  The uncertainty in the CIE weighted irradiance, due to temperature induced errors in the measured solar irradiance spectra, will be examined.
 
MATERIALS AND METHODS

Instrumentation

The spectroradiometer investigated was an Optronic model 752 (Optronic Laboratories, Orlando, FL, USA). The instrument has been used for three years to measure solar UV radiation on a frequent basis. Component part of the instrument is a double monochromator, focal length 100 mm, F number 4, two concave holographic gratings 1200 lines/mm and a photomultiplier detector (PMT), S-20, cooled relative to the surrounding temperature. The temperature of the PMT  was  approximately 14K below the temperature of the monochromator. The entrance optics used was a 1 m. quartz fiber and a teflon cosine diffuser (Bentham Instruments, Reading, UK). The combination of slits used gave a full width half maximum (FWHM) value of 1.0 nm.
The instrument was temperature stabilized with a home made temperature regulated weather enclosure. The air inside the weather enclosure was circulated and dried, and the temperature of the monochromator could be controlled from the outside. Stable temperatures, within ±0.5 °C, could be achieved in the range between 15 and 40°C, dependent of the surrounding temperature. A combination of heating and thermoelectric cooling (AC2000 Marlov Industries) was used to regulated the system. The temperature was continuously monitored  inside the spectroradiometer at three different sites; the air inside the monochromator, in close connection to the monochromator chassis and close by photomultiplier tube. The air temperature inside the monochromator was the reference temperature for all the results. New software has been developed, to operate the OL752 through a standard GPIB interface, and to store the temperature data together with the spectroradiometer data. The instrument was wavelength calibrated with a Hg line at 312.9 nm, using the OL752-150 Dual Calibration and Gain Check Source Module. Absolute calibration was performed with a 1000W tungsten-halogen lamps, type FEL lamp, issued by Eppley Laboratories, traceable to NIST (National Institute of Standards and Technology, USA). The lamp participated in a lamp intercomparison (18) at Swedish National Testing & Research Institute, SP, in May 1996 and agreed within 3% with the certificate. Slit function was measured with a HeCd laser (325 nm) (18). 
 
Sensitivity

Instrument sensitivity was measured at different temperatures, ranging from 20°C to 35°C.  The temperature was raised in steps of 5K and at each level, the instrument was kept stable for three hours to allow all gradients inside the monochromator to stabilize. The instrument was wavelength calibrated at each temperature level,  when completely stabilized. Sensitivity was tested against a stable light source, a quartz halogen lamp (1000W FEL). The lamp current was supplied by a power supply from Eppley Laboratories. An additional home made current control unit in parallel with the lamp circuit compensated for any drift in the current. Lamp current was adjusted automatically each second and kept stable within 0.2 mA at 7.900 A. Lamp voltage was stable within 0.03%. The distance between lamp and diffuser was kept constant. Measurements of the emission spectrum, from 290 - 400 nm, was repeated ten times at each temperature level. The relative response change as a function of temperature was found. 

Wavelength shift

Two different methods were used to quantify the wavelength shift. One of the  methods was  measurements of a Hg lamp in the laboratory with different instrument temperatures. Wavelength calibration was performed  at 21 °C. Temperature was increased in steps of 2.5 K from  20 °C up to 33 °C and as a control, lowered to 25° C and 20°C. Four lines of the lamp (296.7 nm, 312.9 nm, 365.0 nm, 404.7 nm) were scanned from 2 nm below to 2 nm above the maximum with 0.1 nm resolution, totally 102 spectra. The monochromator was operated in a normal horizontal position and not moved during the experiment. Room temperature was kept constant.
In the second method, wavelength shifts were derived from spectral solar irradiance measurements. The Fraunhofer structure in the global solar spectrum was compared with  the structure in an extraterrestrial spectrum, a method described by Huber et. al. (19) and  Slaper et. al.(20). Two independent sets of irradiance measurements were used. The first set was a data set with 2 nm resolution from June 1995. In this set, the monochromator air temperature changed from 17 to 33 °C and back to 17 °C. Wavelength calibration was performed at 21 °C.  The second data set was from a warm day, May 1997, where the monochromator temperature raised from the normal temperature at 30 °C to 33 °C and cooled down to the normal level, 30 °C,  in the afternoon. In this case, wavelength calibration was performed at 30 °C. Spectral resolution was 0.5 nm. The extraterrestrial spectrum (KittPeak) was convoluted with the instrument slit function and adjusted until minimum deviation between the Fraunhofer structure in the extraterrestrial and the ground level spectra were found. This was repeated for several wavelength intervals of 16 nm for the 0.5 nm resolution data, and 40 nm for the 2 nm resolution data. The method has  an uncertainty in the determination of wavelength shift of ±0.02 nm.
 
Error introduced in the calculated CIE weighted irradiance

A modified version of the PSEUDO SPHERICAL  UVSPEC (21)  model, based on discrete ordinate method (22), was used to create spectra for clear sky conditions. UVSPEC was modified to create a set of high resolution (0.05 nm resolution) ground level spectra with a method, outlined by  Mayer et. al. (23). The ground albedo was set to 10%, ozone level at 340 DU, visibility to 120 km and cloud free conditions were used. Atlas3 extraterrestrial spectrum was used to calculate the spectrum from the transmission calculations. Spectra from 0° to 90° zenith were calculated and convoluted with a real slit function to give an simulated measured high resolution ground level spectrum. A systematic wavelength shift, Dl,  was introduced in the modeled spectra, where the shift was assumed constant for all wavelengths. The error caused by sensitivity change was added, using the modeled sensitivity curves as a function of temperature directly.

RESULTS.

Sensitivity

Our Optronic OL752 is normally operated and calibrated at 30°C. This temperature was chosen as the reference temperature in this study. The temperature dependent sensitivity change was slightly wavelength dependent.  The sensitivity, S, was 7.2% higher at 20°C than the reference, at 25 °C 3.8% higher and at 35 °C  5.1% lower, where the values were averaged over wavelength. S(l) was fitted to a model given by:

where T is the temperature in °C. The deviation in sensitivity from the reference temperature (30 °C) were calculated from this model. Fig. 1 shows the percentage deviation in sensitivity where each contour line represents one percent change in sensitivity relative to the reference temperature. Noise due to fluctuations in the dark current during one scan, was averaged out for all wavelengths. This was done by using wavelength running means of s0(l), s1(l) and s2(l). An average sensitivity change in the ultraviolet region from 290 to 400 nm was approximately -0.8 %/K, within the temperature range tested.

Wavelength shift, Hg lamp

The emission lines from a Hg lamp were measured continuously while the instrument was heated from 20°  to 33°C and cooled down back to the starting temperature. The three lines in the 312.9 nm region, will be observed as one single line with a 1.0 nm bandwidth (FWHM) slit function. In Fig. 2 the measured emission spectra are shown for three different spectroradiometer temperatures.

As the temperature increased, the position of the wavelength with maximum intensity, lmax was shifted towards higher wavelengths. This means that the instrument measured radiation at a lower wavelength, than the setting of the monochromator indicated.  This was defined as a positive wavelength shift.  lmax was found in each emission spectrum. The results for the 312.9 nm triplet are shown in Fig. 3 where lmax is plotted against temperature. There is a strong correlation (R=0.97) between the change in lmax and the spectroradiometer temperature. A slight decrease in intensity due to a reduced sensitivity could be observed as the temperature increased (Fig. 2).

When the temperature was lowered from 33°C and back to 20°C, the wavelength shift was reversed and the maximum wavelength was found at the same position as before.  lmax  was fitted to a linear curve lmax =a+bT  by least squares method, where T is the temperature in °C and b is the temperature coefficient, which quantify the wavelength shift caused by temperature change. Similar temperature coefficients were determined from the other Hg lines (Table 1).

Wavelength shift, solar spectrum

Comparison of measured ground based solar spectra and an extraterrestrial spectrum indicated a similar temperature dependent wavelength shift in the instrument. If the instrument shows a positive shift of 0.1 nm, the extraterrestrial spectrum must be shifted -0.1 nm to be able to find the best matching Fraunhofer structure. A wavelength shift algorithm will indicate a shift of -0.1 nm. All spectral shifts from the wavelength shift algorithm were multiplied with -1 to give the same definition of shift coefficients as in the laboratory experiment. Fig. 4 shows this wavelength shift at 376 nm as a function of temperature. There was a strong correlation between wavelength shift and temperature (R=0.99). Shift coefficients could be determined with regression. The temperature coefficient determined for the entire temperature range was 0.084±0.001 nm/K ( Fig. 4, red line). There was a slight change of slope in the wavelength shift, when the temperature passed above 28 °C (Fig. 4). The coefficient determined for temperatures between 28°C and 33°C (Fig. 4, blue line) was 0.065±0.004 nm/K and for 20 °C to  30 °C the temperature coefficient was 0.09±0.002 nm/K.

Temperature coefficients were derived for the whole wavelength range for global irradiance measurements with 2 nm scan resolution and with 0.5 nm scan resolution. Fig. 5 summarizes temperature coefficients found from the mercury lamp in the temperature range  20-30°C  and solar irradiance measurements, range 30-34°C (0.5 nm scan resolution) and in the range 17-33°C ( 2 nm scan resolution).  

For the highest temperature interval investigated (Fig. 5 ), the coefficients were lower than for the whole temperature range (Fig. 5) or for lower temperatures (Fig. 5 ). The effects of temperature changes were less pronounced at longer wavelengths, independent of the temperature range investigated. When the coefficient was determined in temperature intervals from 28°C to 33 °C, () , we got a overlap with the data determined from the 0.5 nm resolution spectra  determined for the same temperature interval. When we used temperature intervals from  20 °C to 30 °C () matching the mercury lamp tests, we got even higher temperature coefficients. The same was true for temperature coefficients determined for temperatures below 24 °C. 
The shift coefficient varied from 0.080 nm/K at 320 nm to 0.065 nm/K at 380 nm for high temperatures, while the wavelength dependence was less for lower temperature ranges. The temperature coefficient at 254 nm was 0.101 nm/K, 0.062 nm/K at 546 nm  and 0.055 nm/K at 632 nm. The average temperature coefficient, independent of wavelength and temperature range, was stipulated to 0.08 nm/K.

 CIE weighted irradiance
 
CIE-weighted irradiances from clear sky modeled spectra, with and without wavelength shift,  were compared. The deviation in percent was found as a function of wavelength shift and solar zenith angle. Fig 6 shows the error in the CIE weighted irradiance for wavelength shifts from -0.6 nm to 0.6 nm  (corresponding to a temperature change of ±7K), solar zenith angles from 10° to 90° and total ozone 340 DU (material and methods).

A wavelength shift towards shorter wavelengths (negative shift) which occurred when the instrument was operated at a lower temperature than the calibration temperature,  would give too high CIE weighted irradiance. A shift of -0.2 nm,  will result in an overestimation of the CIE-weighted irradiance of 3-4%, dependent on the solar zenith angle (Fig. 6). If the instrument temperature increases above the calibration temperature, a positive wavelength shift will occur and the CIE weighted irradiance will be underestimated. The error became less pronounced when the solar elevation was low (solar zenith angles > 65 degrees).

When the temperature of the instrument deviates from the calibration temperature, two kinds of errors occur, caused by wavelength shift and change in sensitivity. These two errors was assumed additive. To quantify the uncertainty in CIE weighted irradiance due to temperature changes, both the wavelength shift coefficients and the sensitivity coefficients were applied to modeled clear sky spectrum. An averaged wavelength shift coefficient of 0.08 nm /K was used for all wavelengths and independent of temperature range. Sensitivity change was calculated using the fitted sensitivity model (Fig. 1) . Both wavelength shift and sensitivity changes were introduced in clear sky spectra, and the error in the CIE weighted irradiance was calculated for temperature changes in the range from 20°C to 35°C.  Fig. 7 shows the total error in the CIE weighted irradiance when the temperature changes from the calibration temperature of 30°C. Contribution from the wavelength shift error and the sensitivity error can be seen. The example in Fig. 7 gives the error for clear sky conditions, solar zenith angle of 40° and  ozone at 340 DU.

The estimated error in the CIE weighted irradiance obtained when the spectroradiometer operates at  20°C  rather than 30°C was more than 20 %. A temperature change from a calibration temperature at 20°C have the same induced error in the CIE-weighted irradiance from a wavelength shift. The error caused by change in the absolute sensitivity will decrease because of the nonlinear decrease in sensitivity (See Fig. 1). Fluctuations from a stabilization temperature at 20°C gives lower total error pr degree change than a stabilization at higher temperatures.

The decreased instrument sensitivity observed,  as the temperature of the spectroradiometer increases (average -0.8 %/K),  is in agreement with properties found in other spectroradiometers (24) The change in sensitivity correlated both with the temperature change inside the monochromator and at the photomultiplier tube (R=-0.992). It is known in other spectroradiometers that temperature influences the filter transmition (10). We are not able to distinguish whether the effect is caused by filters or the PMT from this experiment, due to strong correlation between monochromator temperature and PMT temperature. The effects are not caused by increased dark current, which increases exponentially with the PMT temperature, because every  spectrum is corrected for dark current. For non-temperature stabilized spectroradiometers operating at different temperatures, a set of calibration curves, valid within a certain temperature range, should be available. If the relation between sensitivity and temperature is known for one instrument, only one calibration curve is needed. The effect of the temperature changes can be taken into account when the data are calibrated. Changes in the instrument sensitivity can be observed with a portable lamp unit, for instance the gain-check lamp that Optronic provides. The uncertainty of these lamps are at least 5% or more (25) but temperature drops of more than five degrees should be detectable.

Temperature properties of the material used in the monochromator and the design of the optics within the monochromator is important for the wavelength stability of the instrument. The two gratings in the OL752 double monochromator are directly connected with a rod and some miss alignment of the gratings might occur as the temperature changes within the monochromator. This is a problem for all spectroradiometers using gratings. However, the magnitude of these effects will be more instrument specific. Both the design of the monochromator, focal length and properties of the gratings and the materials will influence. In the OL752 the wavelength shifted with temperature in a biphasic way. A different temperature coefficient could be found in the lower part of the  temperature range (17 °C to 28 °C), compared with the higher interval (28 °C - 33 °C) (Fig. 4) for the whole wavelength range (Fig. 5).  We do not know the reason for this. There is a time delay between when the air inside the monochromator has come to a stable temperature and when all the parts of the instrument have done the same. However such effects should show up as a hysteresis effect when the instrument is heated and lowered back in temperatures. During this time period temperature gradients might occur within the spectroradiometer.  Even if the temperature nearby the monochromator chassis was monitored and found stable (as in our experiments), there might still be temperature gradients within  instrument that is very difficult to detect and will influence the results.  This might be the explanation why the temperature coefficients found from  the Mercury lamp experiments done in the laboratory deviates slightly from those found with from solar irradiance measurements performed outdoors.

Independent of the uncertainty in the absolute level of the shift coefficients, there is a very clear wavelength dependence (Fig. 5). Emission spectra of the Mercury lamp showed the wavelength shifts in the range from 250 nm to 632 nm. The spectral fine structure that were determined from the solar irradiance spectra (Fig. 5, blue and red points), are probably more due to the method used, rather than a real instrument effect. As an example; it is known that the KittPeak spectrum, used as a reference spectrum for the determination of  the wavelength shift coefficients, give some problems in the 325-330 nm region (26).  This causes some of the strange fluctuations in the shift coefficients derived from this spectrum. For the 0.5 nm resolution scans (Fig. 5, blue points) this cause a strange behavior up to 338 nm, while the same effect influences the results up to 348 nm  in the shift determined from the 2 nm resolution data, due to the method used.  For wavelengths below 325 nm ozone absorption will influence the Fraunhofer structure and cause a systematic underestimation of the wavelength shift. The effect of ozone has not been removed in the data analysis. At low solar elevations (solar zenith angels larger than 70°),  the ozone effect becomes more  pronounced. This is the main reason why the 95% confidence intervals are larger for wavelengths below 325 nm.

The relative errors in the measured solar spectra, caused by wavelength shift, are most serious for shorter wavelengths. The error introduced in the CIE weighted irradiance will depend on the relative shape of the spectrum. This effect was illustrated best will modeled spectra (Fig. 6) where the shift coefficient of the Optronic was applied. The estimated error caused by wavelength shifts (Fig. 6) were in agreement with measured spectra (data not shown). These error estimates will depend upon the ozone concentration. For lower ozone concentrations than 340 DU  larger errors will occur. Choosing another action spectrum with a steeper slope than the CIE, for instance the DNA action spectrum, are more sensitive to wavelength shifts and the estimated errors in the biological dose rates will be larger.

To confirm the total error estimates with measured spectra is not an easy experiments, because very stable atmospheric conditions are needed over a long term. Temperature stabilization of the instrument is very time consuming and measurements should be done at a variety of solar zenith angles. Not good enough data were available to experimentally confirm the assumption that errors due to wavelength shift and sensitivity were independent of each other and to confirm the estimated total error in CIE-weighted irradiance due to temperature changes. For a spectroradiometer used without a temperature control unit, which is the case for most instruments used in  biological field experiments, one can of course eliminate the wavelength shift error by doing wavelength calibration before each scan. This is recommended.  The change in sensitivity can be compensated as long as  the temperature change is known and suitable calibration curves are used. In field experiments where the UV doses should be estimated within 10-20 percent, the effect of temperature should be taken into account. The OL752 model shows an excellent behavior at stabile temperatures. The wavelength accuracy of ±0.1 nm can only be achieved when the temperature of the monochromator is stabilized at a given level with an accuracy of ±0.5 degrees.
Intercomparisons reveal a uncertainty in the determined CIE-weighted irradiance from different instruments of 10-15% (18) . Temperature effects may give a significant contribution to some of the observed deviations between the instruments. A continuous monitoring of the monochromator temperature during operation, is necessary to be able to distinguish temperature effects from other sources of error.

Acknowledgments

This research has been funded by the Norwegian Science Foundation and the European Environmental program SUVDAMA. Anders Johnsson is acknowledged for fruitful comments to the manuscript. NSO/Kitt Peak FTS data used here were produced by NSF/NOAO. For further information contact: fhill@noao.edu (internet) 5/15/95 - F. Hill

References

(2) Bojkov.R.D, V.E Fioletov., D.S. Balis, C.S. Zerefos , T.V Kadygrova  and A.M. Shalamajansky (1996) Further ozone decline during the northern hemisphere winter-spring of 1994-1995 and the new record low ozone over Siberia. Geophysical Research Letters Vol. 22 No. 20, pp. 2729-2732

(3)  Chandra S., C. Varotsos and L.E. Flynn (1996) The mid-latitude total ozone trends in the northern hemisphere, Geophysical Research Letters Vol. 23, No. 5, pp. 555-558
 
(4)  Seckmeyer G.,  R.L. McKenzie (1992) Increased ultraviolet radiation in New Zealand (45° S) relative to Germany (48°N), Nature Vol. 359,  No. 6391, pp. 135-137
 
(5) Kerr,J. and T. McElroy (1993) Evidence for large upward trends of ultraviolet-B radiation linked to ozone depletion. Science, Vol. 262, pp. 1032-1034

(6) Diffey,B. (1996) Measurement and Trends of Terrestrial UVB Radiation in Europe. OEMF spa., Milano. ISBN 88-7076-174-6

(7) Caldwell,M.M., A.H.Teramura, M.Tevini, J.F.Bornmann, L.O.Bjorn and G.Kulandaivelu. (1995) Effects of increased solar ultraviolet radiation on terrestrial plants. Ambio, Vol. 24, No. 3, pp. 166-173, ISBN 951-697-412-0

(8)  Josefsson,W.A.P. (1994) Monitoring ultraviolet radiation. In Environmental UV Photobiology (ed. A.R.Young) Plenum Press, New York, ISBN pp. 73-88

(9)  Kostkowski,H.J. (1997) Reliable spectroradiometry.  Spectroradiometry Consulting, Maryland, ISBN 0-9657713-0, pp.493-495.

(10) Bais, A. F., C.S. Zerefos and C.T. McElroy  (1996) Solar UVB measurements with the double- and single- monochromator Brewer ozone spectrophotometers. Geophysical Research Letters, Vol. 23. No. 8. pp. 833-836

(11)  Gardiner B. G. and P. J. Kirsch  ed. (1995) Setting standards for European ultraviolet spectroradiometers. Air pollution reseach report 53. European Commission, Report EUR 16153, ISBN 92-826-5405-2

(12)  Blumthaler, M. and W.Ambach (1986) Messungen der Temperaturkoeffizienten des Robertson-Berger Sunburn Meters and des Eppley UV-radiometers. Arch.Met.Geoph.Biocl., Ser.B Vo. 36, pp. 357-363.

(13) Takeshita S., M. Sasaki, T. Sakata, Y. Miyake and G.A. Zerlaut (1994) Uncertainty of the measurement of global UV-B irradiance using a narrow-band filter radiometer. Proceedings: The eight conference on atmospheric radiation, 23-28 January, Nashville;  Published by the American Meteorological  Society, Boston.  pp. 199-201.

(14) Bernhard G. (1997) Die Genauigkeit von Messungen der solaren UV-Globlstrahlunstärke bei verschiedenen Amosphärenbedinungen, Shriftenreihe des Fraunhofer-Instituts Atmpsphärische Umwltforchung. Band 47. ISBN 3-927548-98-7

(15) Forster P. M. de F., K. P.Shine, A.R. Webb (1995) Modeling Ultraviolet Radiation at the Earth's Surface. Part II. Model and Instrument Comparison., Journal of Applied Meteorology, Vol. 34 pp. 2426-2439

(16) Booth, C.R., T. B. Lucas , J.H. Morrow, C.S. Weiler and P.A. Penhale (1994) The United States National Science Foundation's polar network for monitoring ultraviolet radiation., Ultraviolet radiation in Antarctica: Measurements and biological effects, Antarctic research series. Vol. 62, pp. 17-37.

(17) CIE (International Commission on Illumination) (1987) Research note. A reference action spectrum for ultraviolet induced erythema in human skin. CIE J.6, pp. 17-21

(18)  Koskela T. ed.,   NOGIC93 The Nordic Intercomparison of ultraviolet and total ozone instruments at Izaña from 24 October to 5 November 1993. Final Report. Finish Meteorological Institute, Helsinki 1994
 
(19) Huber M., M. Blumthaler, and W. Ambach (1993) A Method for determining the wavelength shift for measurements of solar UV-radiation. Atmospheric Radiation, (SPIE 2049) pp. 354-357

(20)  Slaper H., H.A.J.M. Reinen and M. Blumthaler and M. Huber and F. Fuik (1995) Comparing ground-level spectrally resolved solar UV measurements using various instruments: A technique  resolving effects of wavelength shift and slit width. Geophysical research Letters. Vol. 20,pp.  2721-2724
 
(21) Kylling A., (1995) 'UVspec, a program package for calculation of diffuse and direct uv and visible intensities and fluxes', available by anonymous ftp to kaja.gi.alaska.edu, cd pub/arve.

(22) Stamnes K., C. S. Tsay, W. Wiscombe and K. Jayaweera  (1988) Numerically Stable algorithm for discrete-ordinate method radiative transfer in multiple scattering and emitting layered media.,  Applied Optics Vol. 27 pp. 2502-2509

(23) B. Mayer,  G. Seckmeyer and A. Kylling., (1997) , Systematic long-term comparison of spectral UV measurements and UVSPEC modeling results. Journal of Geophysical Research, Vol. 102. No. D7. pp. 8755-8767.
 
(24) Svenø, T. Personal communication.

(25) NOGIC96  The Nordic Intercomparison of ultraviolet and total ozone instruments at Izaña from 8.-22. October 1996. Finnish Meteorological Institute, to be published.
 
(26) Slaper H.,  Personal communication.