Version 2 Network Data Corrections
The following corrections have been implemented in Version 2:
- Wavelength correction
- Cosine error correction
- Bandwidth normalization
- Homogenization of wavelength steps
Detailed description of data products:
The wavelength calibration of Version 0 data measured before 1997 is approximately 0.1 nm different from the calibration of later data. This step-change has been corrected in Version 2. The correction procedure is based on a correlation method that compares the Fraunhofer structure in measured spectra with the same structure in a reference spectrum, which has negligible wavelength uncertainty. The actual total ozone column and ozone profile at the time of the measurement are taken into account. Thus, systematic errors caused by the strong increase of the solar spectrum in the UV-B due to ozone absorption and by the wavelength-dependent structure in the ozone absorption cross section (Bass and Paur data) could be minimized.
We estimate that the wavelength uncertainty of Version 2 data is ±0.04 nm (±1σ), independent of the year of measurement. Note: A wavelength error of 0.1 nm causes an error in measured spectral irradiance at 300 nm of approximately 5%. The systematic error in biologically weighted dose-rates due to a 0.1 nm wavelength shift is generally smaller. For example, it is about 4% for DNA-damaging radiation (action spectrum by Setlow, 1974) and 2% for erythemal irradiance.
The signal of a radiometer measuring irradiance should change with the cosine of the angle between the direction of the incident radiation and the normal of the radiometer's collector. All real radiometers, including the SUV-100, deviate from this ideal "cosine-response" to some degree. Version 0 data are not corrected for the "cosine error"; for Version 2 data, a correction was implemented. The correction method is described in the proceedings paper of the conference "Ultraviolet Ground- and Space-based Measurements, Models, and Effects II", which was held in Hangzhou, China, in October 2002. The paper can be downloaded here. A more detailed description is part of a paper accepted by Journal of Geophysical Reserach, which will become available soon.
Before the year 2000, the angular response of SUV-100 spectroradiometers was also depending on wavelength and the solar azimuth angle. An upgrade of the instrument's collectors in 2000 eliminated the azimuth dependence but slightly increased the cosine error. The upgrade introduced a step-change into the Volume 0 time series. The cosine correction method implemented for Version 2 largely reduced this step-change and increased the overall data accuracy.
Version 2 dose-rates for the most common action spectra (such as erythema, DNA-damage, and generalized plant response) are approximately 3-9% higher than Version 0 data. The difference is a complex function, depending on year, time of year (due to the solar zenith angle dependence of the correction), and several other parameters such as cloud cover.
Example of cosine error correction:
o CB940600-351a.pdf: Clear-sky spectrum
o CB940800-349a.pdf: Cloudy-sky spectrum
The ratio of the uncorrected measured spectrum to the complementing model spectrum is indicated by a black line. The same ratio for the corrected spectrum is indicated by red lines. The green line shows the cosine correction, which is independent of wavelength under optically thick clouds (CB940800-349a.pdf). Under clear-skies (CB940600-351a.pdf), the correction has a feature at 505 nm, which is caused by the polarization-dependent throughput of the monochromator ("Wood anomaly").
The bandwidth of SUV-100 spectroradiometers decreases with wavelength. In the UV-B, the full width at half maximum (FWHM) is about 1.0 nm; at 600 nm it is about 0.8 nm. The bandwidth also depends on the specific monochromator that is part of a given instrument. The bandwidth is therefore different for every site and may have changed between years as monochromators have been replaced or refurbished during instrument service. Version 0 data were not corrected for this variation. Version 2 data were normalized to 1.0 nm bandwidth throughout the spectrum. Thus, Version 2 spectra appear as if they were measured with a spectroradiometer with 1.0 nm bandwidth.
The bandwidth normalization facilitates the comparison with model spectra and of spectra that were measured at different sites or during different years at one site. The effect of the normalization on integrals and dose-rates is very small.
Version 0 full-resolution spectra ("composite scans") are given in un-even wavelength steps, which may change from one spectrum to the next. Wavelength increments of Version 2 spectra are uniform. Spectra are given in 0.2 nm steps between 280 and 340 nm, 0.5 nm between 340 and 400 nm, and 1.0 nm above 400 nm. The homogenization greatly helps comparing spectra.
During the first years of network operation, spectra were measured in either 2.5 or 5.0 nm steps above 380 nm. As the bandwidth of the instruments was approximately 0.9 nm in this spectral range, spectra were undersampled. These spectra were re-sampled to a higher resolution by adding the Fraunhofer structure of a reference spectrum.