GOME-2 is a nadir-viewing scanning UV–VIS (visible) spectrometer measuring back-scattered and reflected radiation from the Earth–atmosphere system in a spectral range between 240 and 790 nm . The spectral range is covered by four main optical channels and also by two polarisation measurement devices (PMDs). The polarisation measurements are primarily used to correct for the polarisation sensitivity of the main detectors. The swath width of the GOME-2 scan and the integration time of the channel readouts are programmable. The default 1920 km swath width and the default integration time of 187.5 ms result in 80km×40km (across-track × along-track) ground pixels in the forward scan giving daily coverage at mid- and high latitudes but leaving gaps at low latitudes (Fig. ). To increase the intensity of incoming light, ground pixels with low solar elevation and also the shortest wavelengths subject to strong ozone absorption are measured with a longer integration time. Spectra are also recorded during the backward movement of the scan mirror but the resulting ground pixels are 3 times larger in the across-track direction than the forward scan pixels. A special feature of GOME-2 is that the ground pixel size remains nearly constant over the swath due to a non-linear movement of the scan mirror. In addition to the Earth radiance measurement, a solar irradiance spectrum is measured once a day providing a reference spectrum for atmospheric reflectance.

GOME-2 continues the European contribution to long-term monitoring of atmospheric ozone started by the GOME aboard the second European Remote Sensing satellite (ERS-2; launched in 1995) and the Scanning Imaging Absorption spectroMeter for Atmospheric CartograpHY (SCIAMACHY) aboard the Environmental Satellite (Envisat; launched in 2002). Total ozone columns are produced by the German Aerospace Center (DLR) in the framework of the O3M SAF and disseminated to the near-real-time users via EUMETSAT's Multicast Distribution System (EUMETCast). The operational retrieval is based on the GOME data processor (GDP) version 4 family of algorithms using differential optical absorption spectroscopy (DOAS) . The algorithm consists of a slant column fit of the measured GOME-2 reflectance in a 325–335 nm fitting window covering the Huggins band absorption features of ozone to an equation based on the Lambert–Beer absorption law and including a polynomial closure term to deal with broadband signatures over the fitting window. An iterative air mass factor is used to convert the slant column to vertical column density. The conversion step involves online radiative transfer computations, utilisation of cloud parameters (cloud top height and albedo, effective cloud fraction) obtained in a preprocessing step by an Optical Cloud Recognition Algorithm (OCRA) and Retrieval of Cloud Information using Neural Networks (ROCINN) algorithms , and addition of a climatological ghost column below the cloud level.

Several enhancements to the basic algorithm were introduced in GDP 4.4: improved cloud retrieval algorithms including detection of Sun glint effects, a correction for intracloud ozone, better treatment of snow and ice conditions, accurate radiative transfer modelling for large viewing angles and elimination of scan angle dependencies . GDP 4.7 introduced further improvements over the basic algorithm . Most notably, the ozone absorption cross sections were updated to the Brion–Daumont–Malicet (BDM) data set .

The UV bands of the GOME-2 instrument suffer from severe throughput degradation . Although the DOAS approach is relatively insensitive to instrument degradation, the degradation is evident in the increase of the fitting residuals . A recent validation study, however, indicates that GOME-2 total ozone agrees at the ±1 % level with the standard ground-based Dobson and Brewer measurements and also with corresponding satellite-based data sets from GOME/ERS-2, SCIAMACHY/ENVISAT and OMI/Aura , therefore providing a good quality input data source for radiative transfer modelling of surface UV irradiance.

The AVHRR/3 aboard the Metop satellites is a heritage instrument provided by NOAA. It is the latest version of the series carried on the POES series of satellites, beginning with a four-channel instrument aboard TIROS-N in 1978 . The third version was first carried on NOAA-15 launched in May 1998. The current operational NOAA-19 is the last NOAA satellite carrying AVHRR/3. It has been replaced with the Visible Infrared Imaging Radiometer Suite (VIIRS) in the Suomi-NPP and Joint Polar Satellite System (JPSS) programs. AVHRR/3 is a broad-band six-channel scanner sensing in the visible, near-infrared and thermal infrared spectral regions. The channel 1 is located in the visible at 580–680 nm (Fig. ) with an effective wavelength of ca. 630 nm and an equivalent width of ca. 80 nm, the exact values depending on the individual instrument. The swath width is ca. 2900 km, much wider than for GOME-2, providing a daily global coverage also at low latitudes. The onboard processor stores the measurements in two different data formats with different ground pixel sizes. The local area coverage (LAC) data are stored in full resolution of the scanner, approximately 1.1 km at the satellite nadir point. The global area coverage (GAC) format is a thinned and averaged format. First, data are only used from every third scan line. Second, the amount of data are further reduced in the scan line direction by storing only the average value of four adjacent samples and skipping one sample before moving to the next set of four samples. The area covered by a GAC pixel at nadir is therefore formally 3.3km×5.5km in the along and across track directions, respectively, although only a 1.1km×4.4km fraction contributes to the signal. Both GAC and LAC pixels increase in size towards the edges of the swath.

Unfortunately, there are no onboard calibration devices for the visible channels. The calibration coefficients determined prior to launch are traceable to the radiance standards maintained at the National Institute of Standards and Technology (NIST) . The calibration is known to degrade in orbit and is maintained with vicarious post-launch calibration techniques . EUMETSAT processes the raw instrument data from Metop AVHRR/3 in LAC format and NOAA AVHRR/3 in GAC format to calibrated and geolocated level 1b radiance products, and broadcasts them to the near-real-time users via the EUMETCast system .

The surface height is obtained from the US Geological Survey's Global 30 Arc-Second (GTOPO30) digital elevation model covering the full extent of latitude from 90∘ S to 90∘ N and the full extent of longitude from 180∘ W to 180∘ E at regularly spaced 30 arcsec (ca. 1 km) intervals. The mean surface height together with the minimum and maximum values are computed from the GTOPO30 map for each 0.5∘×0.5∘ grid cell. The surface pressure for each grid cell is then obtained from the hydrostatic equation using a scale height of 7.5 km. In this way synoptic variations in surface pressure (high and low pressure systems) are neglected in the processing. The minimum and maximum height grids are used for quality flagging. If the minimum or maximum height differs from the mean value more than a threshold value, currently set to 750 m, the cell is quality flagged as containing an inhomogeneous surface to indicate that the underlying assumption of a homogeneous horizontal surface (Sect. ) is problematic for these grid cells, typically containing mountain slopes or edges of ice sheets.

In order to reduce dependencies on external inputs, the operational algorithm relies on climatologies for the surface albedo and aerosol optical depth. The surface albedo climatology for the 0.5∘×0.5∘ latitude–longitude grid was constructed for every day of year based on the presence of snow or ice in each grid cell. Northern hemispheric monthly snow cover extent data  from the International Satellite Land-Surface Climatology Project, Initiative II (ISLSCP II)  together with the monthly masks of maximum sea ice extent derived by the National Snow and Ice Data Center (NSIDC) from the sea ice concentrations obtained from passive microwave data  were used to determine the days of year when snow or ice was present in each grid cell. The monthly Minimum Lambert Equivalent Reflectivity (MLER) climatology  is then used for regions and time periods with permanent or negligible snow/ice cover while a climatology better capturing the seasonal changes in the surface albedo during the snow/ice melting and formation periods  is used elsewhere. The surface albedo in the visible wavelength range from ca. 580 to 680 nm covered by the AVHRR/3 channel 1 deviates from the UV albedo, especially at deserts where the soil minerals absorb the UV more strongly than in the visible wavelengths . The difference in UV and visible albedo is accounted for by using the University of Maryland 8 km land cover data  and the dependency of average MLER for different surface types on wavelength, obtained from 5.5 years of GOME observations covering the period June 1995–December 2000 . For example, Fig. 3 in  indicates that over deserts the visible albedo (taken at 610 nm) is 3 times larger than the UV albedo (taken at 380 nm).

Partially ice or snow covered grid cells are inhomogeneous with respect to surface albedo, especially near the edges of ice sheets and around small snow or ice covered islands where one part is water with a low UV reflectivity and the other part is ice or snow with a high UV reflectivity. It is assumed that in order for a grid cell to be homogeneous, the area defined by the current cell and its nearest neighbouring cells has to be homogeneous. The minimum and maximum values of surface albedo for this area are determined, and if the difference between the maximum and minimum is larger than a threshold value, currently set to 0.1, the current grid cell is quality flagged as inhomogeneous to indicate that the surface UV quantities vary significantly within the grid cell due to the surface albedo.

Obtaining reliable aerosol information from satellite measurements is a complicated task . The wavelength-dependent aerosol signal is often mixed with signals from the surface, trace gas absorption, Rayleigh scattering and scattering from sub-pixel clouds. Combining different satellite-derived aerosol products from different sources is error prone in the operational processor, and therefore the aerosol optical depths, τa, at 550 nm are currently obtained from a monthly climatology  combining aerosol products from various satellite instruments covering the time period from 1979 to 2006, including TOMS, AVHRR, POlarization and Directionality of the Earth's Reflectances (POLDER), Multi-angle Imaging SpectroRadiometer (MISR) and Moderate-Resolution Imaging Spectroradiometer (MODIS), to a composite that best agrees with the high quality data from the ground-based sun-photometer measurements in the Aerosol Robotic Network (AERONET) .

The calculation of dose rates is based on the hemispherical spectral irradiance Eh (Wm-2nm-1) incident on a horizontal surface at a given time t, obtained from Eh(λ,τ,t)=∫02πdϕ∫0π2L(λ,θ,ϕ,τ,t)cos⁡θsin⁡θdθ, where θ and ϕ are the zenith and azimuth angles defining the direction of the spectral radiance L (Wm-2sr-1nm-1) incident on the horizontal surface and λ is the wavelength (nm). The atmospheric optical depth τ is time dependent through variation of the cloud optical depth and total ozone column during the day. For simplicity, parameters assumed to be time independent, such as surface pressure, surface albedo, aerosol optical depth and atmospheric profiles, are not explicitly shown. The incident radiance L includes the direct beam from the Sun and is obtained with radiative transfer modelling (Sect. ). It depends on time through the variation of the Sun–Earth distance during the year and solar elevation during the day, and also through τ. Time variations in the solar spectrum are neglected. The hemispherical spectral irradiance is multiplied by a given weighting function w(λ) and integrated over the wavelength λ in a given spectral range [λ0,λ1] to give Ew′(τ,t)=∫λ=λ0λ1w(λ)Eh(λ,τ,t)dλ.

In this paper, we call the weighted and integrated irradiance Ew′ the dose rate (Wm-2). The dose rate is then integrated over the sunlit part of the day to give the daily dose Hw (Jm-2) Hw=∫t=sunrisesunsetEw′(τ,t)dt involving interpolation of the optical depth to the integration time steps as described in Sect. . The daily doses obtained for each weighting function described below are stored in the product. Also stored are the daily maximum dose rates Ew, max′ for each weighting function, given by Ew, max′=arg maxtEw′(τ,t), and the solar noon UV index, obtained by multiplying the erythemal dose rate at solar noon in units Wm-2 by a factor of 40 m2W-1 to obtain a unitless number in a convenient scale .

The erythemal (CIE) weighting function  describes the response of the human skin to UV radiation causing reddening of the skin (i.e. sunburn). It is defined piecewise as w(λ)=1.0 : 250≤λ≤298nm100.094(298-λ) : 298<λ≤328nm100.015(139-λ) : 328<λ≤400nm.

The weighting function for DNA damage  describes the ability of UV irradiance to cause damage to unprotected DNA, and is defined by w(λ)=e13.82(1.0D-1.0)0.0326,D=1.0+eλ-3109, where normalisation to 1.0 at 300 nm is used (normalisation to 265 nm is divided by 0.0326). The weighting function for generalised plant response  is defined by w(λ)=2.6180.21761.0-λ313.32e-λ-30031.08, where normalisation to 1.0 at 300 nm is used (normalisation to 280 nm is divided by 0.2176). Initial versions of the product included the Skin Cancer Utrecht Philadelphia for human beings (SCUP-h) weighting function . In version 1.20, however, it was replaced with the weighting function for the production of previtamin D3 in human skin (in this paper referred to as “vitamin D synthesis” for simplicity), obtained by linear interpolation of the tabulated data published by the . Finally, the weighting function is equal to 1 for integrated UVB and UVA radiation. Figure  (left) shows the different weighting functions currently used in the product.

Two important tropospheric photolysis reactions are driven by UV radiation. For the formation of atomic oxygen in its exited 1D state from ozone O3+hν(λ<320nm)→O(1D)+O2 the rate constant is jO(1D). This is an important photodissociation route of ozone leading to production of hydroxyl radical, a key species in oxidation of hydrocarbons in the troposphere. For the photolysis of nitrogen dioxide NO2+hν(λ<420nm)→NO+O the rate constant is jNO2. The atomic oxygen can react with molecular oxygen producing tropospheric ozone.

For the photolysis frequencies, the spherical spectral irradiance (actinic flux) Es, expressed here in photonss-1m-2nm-1, for a given time is computed from Es(λ,τ,t)=∫02πdϕ∫0πL(λ,θ,ϕ,τ,t)sin⁡θdθ and the photolysis frequencies jX in s-1 are obtained by weighting and integrating over wavelength jX(τ,t)=∫λσX(λ,T)ΦX(λ,T)Es(λ,τ,t)dλ, where σX(λ,T) is the absorption cross section (m2molec-1) and ΦX(λ,T) is the photolysis quantum yield (molecphoton-1) for species X. For the O(1D) reaction, the BDM absorption cross sections are used while the quantum yields are from . For the NO2 reaction, absorption cross sections by  and quantum yields from  are used. The absorption cross sections and quantum yields are plotted in Fig.  (right). The daily maximum jO(1D) at the surface level was introduced in product version 1.20 on 9 July 2013, and the daily maximum jNO2 was included on 15 March 2015.

Cloud optical depth is retrieved from the channel 1 reflectance of AVHRR/3 aboard Metop and NOAA satellites. One satellite, nominally the prime, from both programmes is used at any given time. Figure  shows the average number of cloud observations obtained from AVHRR/3 using the operational combination of Metop-A with either NOAA-18 or NOAA-19 during the period from 1 June 2007 to 31 December 2012. As expected, one cloud observation from each satellite is obtained at low latitudes while more than five are obtained at high latitudes in the summer.

The cloud optical depth is retrieved in the lowest common spatial resolution, i.e. the GAC resolution of the NOAA data, and therefore the Metop LAC-resolution data are thinned and averaged to the GAC resolution prior to processing. The L1/E0,1 ratios are computed from the radiance and solar irradiance given in the level 1b data file, and the cloud optical depth is interpolated from the look-up table for each GAC pixel. The measurement is discarded if its value is out of the range of the look-up table. For each overpass of a given grid cell, the cloud optical depths of GAC pixels hitting the cell are averaged to form a grid cell average observation at an average overpass time. Taking the average is consistent with the assumption of a homogeneous cloud layer (Sect. ), but due to the non-linear dependence of the reflectance on cloud optical depth (Fig. ) leads to an error in cloudiness situations involving averaging over very large cloud optical depth differences, such as partial cloudiness. In future versions of the algorithm, the high spatial resolution of the Metop LAC data can be exploited to improve the modelling of partial cloudiness.

The AVHRR/3 ground pixels at the GAC resolution are smaller than the 0.5∘×0.5∘ grid cell used for the surface height and albedo grids. In the cloud optical depth retrieval it is assumed that the grid cell is sufficiently homogeneous with respect to the surface properties to allow for using the 0.5∘×0.5∘ surface property grids for the GAC pixels. As discussed in Sect. , a quality flag is set if the surface properties are too inhomogeneous according to the predefined criteria.

The sunlit part of the diurnal cycle is discretised with a half-hour time steps from sunrise to sunset taken as the times when the solar zenith angle is 88∘, the largest value in the dose rate and photolysis frequency look-up table. The grid average total ozone and cloud optical depth observations Oovp for each overpass time are interpolated to the times of the diurnal cycle using the nearest neighbour interpolation, represented by O(td)=Oovp(iovp), where iovp is the index of the overpass closest in time to the diurnal time td. This approach avoids the possibly large errors in the observations measured with large solar zenith angles early in the morning or late in the evening affecting the important high dose rate region near noon. For cloud optical depth observations retrieved only for θ0<70∘ (Sect. ), this means that a constant value for τc is assumed between 70∘<θ0<88∘, although the value can be different in the morning and evening. If no cloud observations are available, the grid cell is flagged as missing data, except on the Greenland and Antarctica ice sheets where cloud-free conditions are always assumed (Sect. ). The dose rates and photolysis frequencies are then interpolated from the look-up table using the interpolated ozone and cloud optical depth values. The aerosol optical depth and surface albedo are constant for a given day and taken from climatologies as described in Sect. . The diurnal integral given by Eq. () is evaluated using trapezoidal integration over the half-hourly values. The daily maximum values are computed from Eq. () and the solar noon UV index is obtained from the diurnal data point referring to the solar noon. The values are stored in a HDF5 file as described in the product user manual .

Figure  compares the diurnal erythemal dose rates extracted from the product processing with ground-based measurements using a UVB Biometer model 501 radiometer from Solar Light Co. (SL-501) in Sodankylä (67.37∘ N, 26.63∘ E) on selected days in June 2007 with different cloud cover. On the fairly clear day of 2 June, the modelled and the measured dose rates matched well. As expected, more deviations are seen on the cloudy days of 3, 6 and 20 June because the spatial and temporal sensitivity of the two methods are different. The ground-based values are measured every 10 min on a very confined area, therefore reacting to every single piece of cloud passing over the measurement site. The satellite data, on the other hand, represent half-hourly averages in the 0.5∘×0.5∘ grid cell (ca. 55km×21km at this latitude). Despite these differences in the two data sets, the overall shape of the diurnal cycle is captured demonstrating the usefulness of multiple overpasses during the day.

A number of quality flags are set during the product processing to indicate the expected quality for each grid cell. The flags related to surface homogeneity were already discussed in Sect. . Clouds are a major source of error, and therefore several flags related to clouds are shortly discussed here. Figure  shows the dependence of the erythemal dose rate on the channel 1 reflectance for different surface albedos derived from data shown in Figs.  and . Also shown is the absolute value of the elasticity ϵ of the dependence ϵ=R1dEery′Eery′dR1, giving the percent change in the erythemal dose rate with respect to a percent change in the channel 1 reflectance. At large reflectances the elasticity increases rapidly with reflectance and is ca. 15 % at a reflectance of 0.9. Even after the vicarious in-orbit calibration procedures, the error in the AVHRR/3 channel 1 reflectance tends to exceed the 1 % level  and the error in the erythemal dose rate becomes too large at this limiting reflectance. The exact reflectance limit depends on measurement geometry and atmospheric assumptions (not shown). Currently a limiting cloud optical depth of 80 is used to set a quality flag warning for thick clouds (Fig. ). On the other hand, at a very high surface albedo close to unity no cloud optical depth can be retrieved from the channel 1 reflectance as indicated by the close-to-zero derivative in Fig.  and the abrupt behaviour of the elasticity in Fig. . In the interior parts of the Greenland and Antarctica ice sheets; however, typical clouds are optically thin (τc<1) , and therefore cloud-free conditions are assumed for these regions. This is seen in Fig.  where the ice sheets have valid values beyond the limiting solar zenith angle of 70∘ set by the cloud optical depth retrieval. The limiting values of the quality flag settings are subject to change during the evolution of the processor and up-to-date values can be found in the product user manual  and the product specific metadata section of the HDF5 file.

Overall consistency of the latest processed product with the existing data record is verified by monitoring the global average value of the erythemal daily dose. The monitoring results are shown on an online quality monitoring page at the O3M SAF web site. An intercomparison of the product with traditional ground-based surface UV measurements will be presented in a separate paper.