3.2.5. Broadband UV
A broadband UV instrument (Yankee UVB-1, SN 950208) has been in operation at MLO since August 1995. Additional broadband UV instruments are being used routinely at BRW, BAO, Bermuda, and Kwajalein as part of the BSRN program. These additional instruments have all been compared at various times with the MLO instrument. A calibration procedure was developed for the broadband instruments using a long-term comparison (1 year) between the MLO UVB-1 and the MLO UV spectroradiometer that was installed at MLO in July 1995 [Bodhaine et al., 1998].
One year of clear sky data was analyzed for SZA of 5°-85°, in steps of 5° and for total ozone values in the range 220-310 DU measured with a Dobson spectrophotometer. Because the erythemal response defined for human skin is significantly different than that of the broadband instrument, the calibration of the broadband instrument reporting in erythemal units is strongly dependent on total ozone. When a broadband instrument is placed in the field, it is necessary to know the calibration as a function of ozone to determine accurate erythemal irradiance. However, the manufacturers of broadband instruments do not generally provide information on the ozone dependence of the calibration. The procedure described here provides this information and does not require precise knowledge of the spectral response of the broadband instrument.
Instrumentation
The principle of operation of the UVB-1 instrument depends on a UV-sensitive phosphor that absorbs radiation in the UV-B region and re-emits in the green region. Radiation from the whole sky passes through a quartz dome and is incident on a horizontal UV broadband filter that transmits the UV to the phosphor. The green light emitted by the phosphor passes through a green filter and is measured by a photodiode that has its peak response in the green part of the spectrum. The UVB-1 is temperature stabilized at 45°C.
The manufacturer provides conversion factors for estimating various portions or weighted integrals of the UV spectrum, such as total UVB (280-315 nm or 280-320 nm), Diffey Action Spectrum, Parrish Action Spectrum, or the DNA weighted spectrum. However, these estimates can be significantly in error because of the fact that the actual spectral response of the instrument can be significantly different than the portion of the spectrum being estimated which in turn causes strong ozone dependence of the measurements. This occurs because of the strong spectral variation of ozone absorption of UV in the UVB region of the spectrum. The absolute calibration of Robertson Berger (RB)-type meters, including detailed theoretical studies of these instruments, was treated by DeLuisi and Harris [1981, 1983] and more recently by Grainger et al. [1993]; Mayer and Seckmeyer [1996]; and Leszczynski et al. [1998]. All of these authors compared broadband instruments with a collocated spectroradio-meter.
The UV spectroradiometer located at MLO and used in this study was described by McKenzie et al. [1992] and its operation at MLO was described by Bodhaine et al. [1997]. This instrument uses a horizontally mounted diffuser designed to view the whole sky and minimize cosine error. The spectroradiometer is programmed to begin measurements at dawn and perform scans at 5° SZA intervals throughout the day beginning and ending at 95°, except for a period of time during the middle of the day when the system switches to a scan every 15 minutes.
The UVB-1 data set consists of 3-min means obtained from the MLO data acquisition system. The spectral response of the UVB-1 was taken from the calibration sheet supplied by the manufacturer for this particular instrument, and is shown in Figure 3.20. Also shown in Figure 3.20 is the International Commission on the Environment (CIE) erythema weighting function provided by McKinlay and Diffey [1987] to represent the response of human skin to UV irradiance. The two spectra in Figure 3.20 were used to construct two data sets from the spectroradiometer data by weighting and integrating over wavelength. The three data sets used here are:
UVB1: Broadband instrument voltages (V),
S(CIE): CIE spectrum weighted spectroradiometer (mW cm-2),
S(UVB1): UVB-1 spectrum weighted spectroradiometer (mW cm-2).
Fig. 3.20. Spectral response of the UVB-1 SN950208 (manufacturers specifications) compared to the CIE approved spectrum for the response of human skin.
Three comparisons were done. UVB1 and S(CIE) were compared directly to derive the UVB-1 instrument calibration. The UVB1 and S(UVB1) data sets were compared to show the expected minimum effect of ozone. Also, S(UVB1) and S(CIE) were compared to show a method of simulating the different responses of the UVB-1 and the CIE spectra using spectroradiometer data.
Data Analysis
For the following analysis, UV spectroradiometer data for SZA ³ 45° were chosen for clear mornings at MLO during the July 1995 to July 1996 time period. This gives 1 full year of data, amounting to 132 data points, allowing the study of an annual cycle, and giving ozone values in the range 220-310 DU. A day was accepted as a clear day as described in the previous section. Data for SZAs smaller than 45° (sun higher in the sky) were obtained from days when morning clear conditions extended long enough, and during those times of the year when the sun was high enough in the sky. All Dobson ozone data were derived from A-D direct sun ground quartz plate observations [Komhyr et al., 1993].
In order to ensure that the spectroradiometer and UVB-1 data were simultaneous, a linear interpolation was performed on the UVB-1 voltage time series in order to assign values coincident in time with spectroradiometer weighted irradiance values. In this way the three data sets could be compared at the same effective times. Finally, the three ratios (UVB1)/S(CIE), (UVB1)/S(UVB1), and S(UVB1)/S(CIE) were formed in preparation for regression against the ozone data set.
Figure 3.21b shows the ratio (UVB1)/S(CIE) plotted as a time series for the SZA 45° data. If the instruments were in exact agreement, the ratio would be equal to 1 everywhere. However, the time series show not only an offset but also an annual cycle that appears to correlate very well with the MLO ozone time series shown in Figure 3.21a. This correlation with ozone suggests that the calibration of the broadband instrument depends significantly on total ozone.
Fig. 3.21. Time series of (a) total ozone (Dobson), and (b) ratio of UVB-1 erythema to spectroradiometer erythema (CIE-weighted spectra) at SZA 45°. UVB-1 erythema values were calculated using the manufacturers suggested calibration factor (0.141 W cm-2 V-1). Note that the average of the erythema ratios appears to be close to the slope of the regression line in Fig. 2.
On the basis of the apparent dependence of the ratio (UVB1)/S(CIE) on total ozone, the following procedure was developed for the calibration of the UVB-1 broadband instrument in CIE erythema units. An example for SZA 45° is shown in Figure 3.22. The ratio (UVB1)/S(CIE) was regressed against the ozone data set to produce a linear equation describing the relationship. Some scatter exists about the regression line but the correlation is still fairly high. The equation relating the ratio (UVB1)/S(CIE) and ozone was then used to recover estimated erythema values, given UVB-1 voltages at the prescribed SZA (45° in this case), where total ozone is the independent variable. This process was repeated for SZAs from 5° to 85° in steps of 5°, giving a family of linear calibration equations depending only on ozone. The regression equations were expanded into an array of calibration factors depending on SZA and ozone and the results are shown in Figure 3.23 as a family of curves where different curves are given for different values of total ozone. This revealing figure shows that the ozone dependence is positive for small SZAs, becomes zero at SZA 65°, and becomes negative at large SZAs. This strange behavior is because different spectra respond differently to ozone. This effect is explained in more detail by Bodhaine et al. [1998].
Fig. 3.22. Calibration analysis of the UVB-1 broadband instrument voltages compared to the CIE-weighted spectroradiometer data (erythema) as a function of ozone for SZA 45°. The graph shows the regression of the ratio (UVB1)/S(CIE) against ozone. Note the strong dependence on ozone.
Fig. 3.23. Family of calibration factor curves as a function of SZA for various values of ozone. The calibration factor is essentially S(CIE)/(UVB1) so that multiplying the UVB-1 voltage by the calibration factor for the proper SZA and ozone value gives the predicted erythema for the broadband instrument.
To obtain a more complete understanding of the effects discussed here, two additional similar sets of analyses were performed. To show that the dependence on ozone is because of the different spectral responses of UVB1 and CIE, the entire analysis was repeated using the UVB1 and S(UVB1) data sets that presumably have the same spectral response. The ozone dependence was found to be small in this case [see Bodhaine et al., 1998].
In the third case study S(CIE) and S(UVB1) were compared. Since the same sensor was used for these two data sets by treating the ratio S(UVB1)/S(CIE), instrumental effects are minimized. Thus this test should reveal the true ozone dependence that results from comparing the two different spectra. Figure 3.24 shows the family of curves that gives the correction factor as a function of SZA for various values of ozone. In this case the ozone dependence is almost wholly due to the two different spectra and is not significantly influenced by the use of two different sensors.
Fig. 3.24. Family of calibration factor curves as a function of SZA for various values of ozone. The calibration factor is essentially S(CIE)/S(UVB1) so that multiplying the S(UVB1) data by the calibration factor for the proper SZA and ozone value gives the predicted erythema for the broadband instrument.
Model Results
To investigate the results of the instrument comparison given previously, radiative transfer (RT) calculations simulating the above measurements were performed. Solar UV irradiances with spectral resolution similar to that of the MLO spectroradiometer were calculated for MLO atmospheric conditions using a version of DISORT [Stamnes et al., 1988] that was implemented at CMDL. The models spectral results for various SZAs and ozone amounts were weighted by the S(CIE) and S(UVB1) spectra to simulate the measurements given in Figure 3.24, and the results of the model calculations are shown in Figure 3.25. It is seen that the model calculations agree with the measurements in Figure 3.24 within about 7% and the general shape of the curves is replicated. Model calculations of a somewhat similar nature were performed by Leszczynski et al. [1998].
Fig. 3.25. Model calculations of the family of calibration factor curves as a function of SZA for various values of ozone using the CMDL version of DISORT to simulate S(CIE)/S(UVB1). This figure may be compared directly with Figure 3.26. Details of the model are given in the text.
Because broadband UV sensors, such as the RB type instrument, are commonly used to estimate erythemal irradiance, it is extremely important to develop a calibration procedure applicable under a full range of field conditions. The best method of calibrating a broadband sensor is to compare it directly with a well-calibrated spectroradiometer and to integrate the resulting spectra over the same spectral response as the broadband instrument. However, a given calibration factor is only good for a particular value of SZA and a particular value of total atmospheric ozone.
This analysis shows that if, for example, total ozone concentration decreased from 300 DU to 200 DU, the calibration constant of a broadband instrument should be increased by almost 20%. Therefore, the broadband instrument would significantly underestimate the increase of erythema.