Dear Dr. Marais,
As you may found, there are still some remaining issues. Please carefully consider them in the revised manuscript.
Best regards,
Yafang Cheng

------------------------------------------------------------------------
1. line 60: Bucsela et al. (2010) should appear on the previous line.

2. lines 99 -100: I still do not understand how taking the difference between the column covering 380 hPa and the tropopause and that covering 380 to 500 hPa yields the column centered at 380 hPa (330 - 450 hPa).

3. line 107: The criterion of cloud radiance fraction > 0.7 allows some amount of signal from the lower troposphere to enter the cloud-sliced data set. Both Choi et al. (2014) and Pickering et al. (2016) favored CRF > 0.9 for this reason. Pickering et al. (2106) report a 5% high bias when using CRF >0.7 compared with CRF > 0.9. A paragraph is needed discussing all of the uncertainties involved in the OMI cloud-sliced NO2 product including this possible bias.

4. lines 257 - 262: The production rate derived here is very model dependent. The authors need to state this fact and that the values derived here are influenced by any error that exists in the UT NOy chemistry that may be in GEOS-Chem. Such possible errors have recently been discussed by Travis et al. (2016) and Silvern et al. (2018).

5. page 3, line 97: Is aerosol really accounted for in the NO2 air mass factors, and if so, how?

Yes. We now reference Boersma et al. (2004) (line 122) that is in turn referenced by Belmonte-Rivas et al. (2015) when describing the air mass factor calculation.

I disagree. Please have another look at Boersma et al., 2004 last sentence of section 3.2. Aerosols are not explicitly taken into account in the DOMINO NO2 data product, only implicitly in partially cloudy scenes through the cloud correction applied. However, this correction has no effect on the cloudy scenes discussed here.

6. page 7, line 199: I understand that aircraft measurements are screened for stratospheric air masses in tropospheric applications. However, here data are compared to satellite retrievals, and these will -­ as far as I understand – include such stratospheric air masses if they are in the right pressure range above a cloud. I therefore wonder if this screening really makes sense here.

Both the aircraft and satellite products exclude stratospheric contributions, and so the comparison is consistent. We now clarify that the OMI satellite product is tropospheric NO2 only (line 74). We already state that the KNMI product removes stratospheric NO2 by subtracting the stratospheric contribution in the retrieval (lines 121-122). We now elaborate that the stratospheric contribution in the NASA product is removed when differencing two nearby partial columns, as NO2 aloft is assumed uniform (lines 134-135 and lines 141-142).

This is not the point. The NASA product takes the difference in NO2 columns over neighbouring clouds having different cloud top. Any NO2 in between these altitudes will contribute to the retrieved UT mixing ratio, regardless of its origin. If stratospheric NO2 is brought down to the UT, the cloud slicing method will detect it as UT NO2 (as it should).

7. Please add the uncertainty estimates for the NOx yield and global lightning NOx source given in the text also to the abstract and conclusions.

----------------------------------------------------------------------