Thank you very much for your replies to the reviewers' comments and for the revised manuscript which you have uploaded as a supplement to the responses. Although uploading a revised manuscript is not required at this stage, I nevertheless think that it is very useful.

Both reviewers agreed that comparing the two different methods to obtain 3-dimensional wet refractivity fields is in principle a good idea, but they both indicated that major revisions would be required before the manuscript can be published. One of the reviewers has indicated a willingness to review the revised manuscript. I therefore invite you to submit a revised manuscript via our manuscript submission system. Based on your responses to the reviewers' comments and the revised manuscript in the supplement, I still have a few questions and comments. I would appreciate if you could take them into account.

One thing that I still ask myself is, how sensitive are the results to details of the methods? While I understand that one cannot address all possible influences in the framework of sensitivity experiments, I think that some more general discussion on this topic would be useful.

In particular, I share some of the concerns of reviewer #2 regarding the WRF setup which in my opinion requires some additional discussion beyond your initial replies. Regarding reviewer #2, Question 6: I think what might be meant here is that one could for example assimilate wind. Water vapor could then still be used for evaluation. I also do not understand why 3DVAR was applied only once at the beginning of the period of interest. Regarding reviewer #2, Question 7: since not only WRFDA is used, but also WRF is run, I do not understand why you "don’t need the model to develop its state in time". If you are using WRF output and the domain is too small to represent the relevant scales, in my opinion, this could very well be a major problem. In principle, I think that it would be good to find out how sensitive the results are to the domain layout (size, position, and resolution). Furthermore, WRF provides a large choice of physics parameterizations, and choosing specific parameterizations could also impact the results. Some discussion on the potential sensitivities of the tomography method to some of the methodological choices might also be beneficial.

I think that at least some of these points should be mentioned in a discussion that addresses the influence of the setup on this comparison.

l. 3, l. 34, l. 58, and elsewhere : filed -> field (and also fileds -> fields)

l. 100: And the start time of the WRF model is epoch of interest. -> please be more specific

l. 171: NCL: NCAR Command Language, I suggest to add a reference (see e.g. https://www.ncl.ucar.edu/citation.shtml)

l. 180: I think WRF output1 and WRF output2 would be more accurate than reanalysis1, reanalysis2.