|Re-review ‘Evaluation of a regional air quality model using satellite column NO2: treatment of|
observation errors and model boundary conditions and emissions’ by R. J. Pope et al.
Many issues raised in my previous review have been addressed in this version of the manuscript by Pope and co-workers. Unfortunately, the section on averaging kernels and the averaged retrieval error have only become more confusing. These parts need to be clarified, or the paper will propagate wrong information in the scientific community.
P6, lines 19-20: kernels do not modify vertical profiles. The proper definition of the averaging kernel for DOAS retrievals is that they represent the relationship between the vertical distribution of the tracer and the retrieved trace column. If kernels are applied on the modeled profile, then this ensures that the resulting modeled column can be compared directly to the satellite-retrieved column. This may all seem a matter of semantics, but it is essential to express oneself clearly in this matter to avoid confusion to proliferate in the literature.
P8, L12-24: the added discussion on why kernels are lower over London than over Dartmoor does not clarify things. I agree with the authors that the kernel differences are probably not driven by differences viewing geometries, albedo, or clouds, and indeed, differences in NO2 loading indeed appear to drive the differences in the kernels. But referring to optical thickess here makes no sense – all the averaging kernels in the DOMINO v2 retrievals have been determined assuming NO2 absorption is sufficiently small so that the optically thin limit may be used. I urge the authors to remove L19-21 (“Eskes and Boersma … of the tracer.”). The mathematical expression of the averaging kernel is the altitude dependent air mass factors divided by the air mass factor. It is the latter that depends on the local NO2 amount (because the AMF depends on the assumed NO2 profile; the altitude dependent AMFs, scattering weights, etc. are always calculated within the optically thin limit). Since (TM4) NO2 amounts are higher over London than over Dartmoor, differences in NO2 loading can indeed explain the differences in kernels. Also in line 22, the reference to optically thin and strong absorbers should be dropped.
P18: the solar diffuser issue was relevant for the GOME-sensor, but OMI does not have that problem. Should be removed.
The main error in the manuscript is with Eq. (6). In my previous review, I remarked that the contribution of the systematic error in the slant column to the (systematic) error in the stratospheric slant column could be discarded in Eq. (6), in line with the discussion in Belmonte-Rivas et al. (2014). This does not mean however, that the stratospheric *uncertainty * can be discarded from Eq. (4). Only the systematic component in the stratospheric can be neglected (0.03 Xtotal/AMFtrop)^2 term), but the random contribution (given by 0.25 x 1015/AMFtrop) should remain, but as the middle term in Eq. (4), and not squared together with the AMF uncertainty term as in the current Eq. (6). The authors should really revise this issue. Failure to do so would propagate an important conceptual error in the peer-reviewed literature, and this should be avoided at all costs.