This gap addresses three of the major individual issues in our understanding of the analysis processing chain from the raw spectrum to the final total column ozone data product using the DOAS technique, and the interpretation of the actual final product.

The first aspect is the uncertainties in the ozone slant columns retrieved with the standard DOAS data analysis fitting procedures. They are to a large part caused by (1) instrumental imperfections such as detector noise, resolution change, etaloning (a fault that develops in thin charge-coupled devices when they behave as etalons) and other non-linearities of the detector, stray-light, and polarisation effects, as well as (2) by issues introduced within the analysis routine such as uncertainties in the Ring effect, unknown absorbers, and the wavelengths dependency of the AMF. Such uncertainties are mostly random in nature and therefore can be estimated statistically from the least-squares fit procedure. However, the fitting uncertainties derived from the least-squares analysis typically result in unrealistically small uncertainties and can lead to an underestimate of the measurement uncertainty by up to a factor of two. Results from intercomparison exercises (e.g. Van Roozendael et al., 1998, Vandaele et al., 2005, Roscoe et al., 2010) show that state-of-the-art instruments hardly ever agree to better than a few percent, even when standardised analysis procedures are used. This indicates that the actual accuracy in the ozone slant columns is at least to some degree limited by uncontrolled instrumental and/or analysis factors. And it leads to the question if something is not yet adequately addressed in the fitting procedures.

Further uncertainties are introduced during the calculation of air mass factors (AMFs) which are required to convert the measured ozone slant columns into vertical columns which means that the measured slant column density (SCD) is divided by the AMF to calculate the vertical column density (VCD) in molecules/cm2 which is then converted into Dobson Units. The NDACC UV-visible spectroscopy working group recommends the use of a generic look-up table of ozone AMFs which has been developed at BIRA-IASB (see NDACC UV-vis working group report) and accounts for the latitudinal and seasonal dependencies of the ozone vertical profiles. The NDACC recommendation is furthermore to average all retrieved vertical columns of ozone between 86° and 91° Solar Zenith Angle (SZA). The recommended approach is to apply a linear fit on vertical columns in the above SZA range and then derive the column value at the effective SZA (so far recommended to be 90° SZA). This range minimizes the measurement uncertainties arising during the fitting procedures and AMF calculation, and provides stratospheric ozone measurements with limited sensitivity to tropospheric ozone and clouds. Ozone and pressure/temperature a priori profiles are key input parameters for the AMF calculations, and AMF uncertainties for zenith-sky twilight ozone retrievals are dominated by uncertainties in a priori profile shape effects. Hendrick et al. (2011) found that the uncertainty in the calculated AMFs based on uncertainties in the ozone profiles is around 1%. However, there is a lack of an adequate database of tropospheric ozone in particular, and in regions where tropospheric or stratospheric ozone contents deviate from the climatological values, uncertainties of several percent can be introduced in total column ozone retrievals. Apart from uncertainties in the ozone a priori profiles, further sources of uncertainty are based on uncertainties in the aerosol and cloud information used. The typically small impact of clouds on zenith-sky ozone UV-vis measurements at twilight is due to the fact that the mean scattering layer is generally located at higher altitude than that of the clouds. However, AMFs calculated for cloudy conditions can be systematically larger than AMFs calculated for non-cloudy conditions.

The DOAS ozone total column retrieval is implicitly dependent on an a priori tracer profile. The radiative transfer calculation within the DOAS analysis accounts for the sensitivity of the measurement to tracer concentrations at all altitudes. These sensitivities are implicitly weighted with the assumed tracer profile to produce the retrieved column.The averaging kernel is proportional to this measurement sensitivity profile, and provides the relation between the retrieved quantities and the true tracer profile. The kernel therefore provides important information needed for a quantitative analysis of the satellite data (Eskes and Boersma, 2003 and references therein). The averaging kernel concept is by now well established in remote sensing. Applications are for instance the retrieval of profiles of atmospheric quantities like temperature and tracers like ozone from satellite measurements. Retrieval groups are increasingly including the kernel information in the profile data products disseminated to users. The look-up tables for total column ozone averaging kernels, provided by the NDACC UV-vis working group, have been developed based on the approach described by Eskes and Boersma (2003), i.e. the averaging kernel of a layer i can be approximated by the ratio of the box airmass factor of this layer i and the total airmass factor calculated from an O3 profile climatology. The availability of averaging kernel information as part of the total column retrieval product is important for the interpretation of the observations, and for applications like chemical data assimilation and detailed satellite validation studies. However, vertical averaging kernels (when provided based on a climatology) are only approximations of the real 3D averaging kernel of a retrieval and cannot fully account for the representativeness of the data.