General Comments:

A nice manuscript/study, fusing two satellite observation techniques to derive an improved temperature, water vapour profile information in the lower troposphere. A few suggestions for improved readability are below. 

Specific Comments:

 L10: “…characterize the complex thermodynamic structures in the lower troposphere” seems to imply this is presented in this study, but the study just proposes a way to improve the characterisation. Please re-phrase. 

L53: Metop-A is no longer flying (or better, providing data), but Metop-B, -C does. So maybe just state all Metops? 

L68: “...they remain challenging to apply in practice” I think one major issue here is that SI traceability is lost when applying such an ad-hoc correction. Maybe that should be pointed out here too. 

L121: Maybe I missed it, but was the symbol e formally introduced? 

L164: “… state vector spanned from 0 to 10 km altitude” Suggest to add that you are ignoring the upper atmosphere in your setup, as the focus is on the lower few km and the contribution of the upper atmosphere decreases exponentially. Maybe even add an uncertainty estimate here.

L171: Are you using 12 or 22 channels in your MWR BT? At L154 it appears the 22 were reduced. 

L194: Just to note that ROPP includes this differentiation, thus no need for time consuming numerical one (but of course does not have the bending angle ducting modifications included).

L229: “…to the a-priori T error with < 1 K difference to the truth” Suggest to state “… error at maximum not even 1K difference…”

L234: Please add the figure you are talking about (2e?)

L236: “…the 1DVar solution for MWR-only tends to follow the shape of the given a-priori” Suggest to add figure being discussed. And I am unsure if the “tends to follow” really captures what the MWR is showing. Seems to more show the same structure. 

Figure 1: Left plot does not show apriori, likely covered by the green curve. And maybe add a full title, and also the figure letters a, b, c, d? As the caption talks about these, but they show nowhere on the plots. And this title point is general for all figures. 

L265: What is this “(magsondewnpnM1.b1.20121104.120900)” exactly? A profile identifier? A file name? If name, maybe better point to where it is available. 

L287: “In all cases, the added observation have a positive impact…” Add weight after observation? There are no new observations added.

L320: “(2020-04-01-03:10c2f4_gps58)” As above.

Figure 6: c,d x labels not really readable. 

Figure 8: The BT shown here at 31.4GHz appear not related to the channels proposed for use here (see Section 2.2: For ATMS, we can focus on channels 4 to 9 (51.76 GHz - 55.5 GHz) that are most sensitive to the tropospheric temperature, and channels 17 to 22 (165.5 GHz to 183.31 GHz) that are most sensitive to water vapor (Shao et al, 2021).) This channel shown appears to be a window one. 

Editorial:

L79: “the the”

L259: “, While”

L284: “Overall, The”

This paper requires major revision. The authors present  a 1D-Var retrieval that combines GNSS-RO bending angles and ATMS radiances with background information. A key result is that combining these two measurement techniques is better than either individually. This result is to be expected. ATMS radiances tend to provide more temperature information in the troposphere and the authors state they can "anchor the solution" (line 118). I think this is misleading because in practice the ATMS radiances will be bias corrected, to account for systematic observation and forward model errors. The calibration/bias correction of the radiances used in the 1D-Var retrieval needs to be discussed in some detail. In addition, Collard and Healy (2003) demonstrated that RO temperature information falls as we move closer to the surface, so does this work tell us anything new about the information content of RO measurements?

The simulations appear technically correct, but the 1D-Var B matrix seems unrealistic. The background errors will be vertically correlated. A 2.5 K uncertainty (line 183) may have been appropriate in 2003, but it is not now. Similarly a -2 K bias in the a priori over the entire profile  (line 215) is not realistic. A short-range forecast from NCEP, the Met Office,  ... would not contain a bias of this magnitude. Therefore, assessing the value of observations on this basis does not seem reasonable. 

On line 368 it states that the retrieval is "independent of any operational NWP model", but on line 329 it says that "the NCEP analysis, used as priors in the 1DVar". This needs to be clarified. In addition, did the NCEP analysis assimilate COSMIC-2 and ATMS? On line 361, the authors rightly note the similarities with data assimilation for numerical weather prediction, which is designed to retrieve information from a broad range of observation types. On 369, it states that the joint retrieval may be useful for validating NWP models. However, it is difficult to believe that is joint retrieval will be more accurate than an NWP analysis where COSMIC-2, ATMS and other observations have been assimilated. The authors will have to justify this potential NWP application more clearly before publication.