Please see the attached comments. The figure exceeded the system’s upload limit, so I’ve included it along with the comments in the attached file.

Review of “Continuing the MLS water vapor record with OMPS LP”  by Himes et al.

This manuscript presents a new water vapor dataset derived from OMPS LP radiances using a neural network trained on collocated MLS measurements. While the OMPS LP water vapor show promise, there are still some major concerns to be considered, as discussed below.

My main concern is that “when omitting 2024 during training, the model begins producing severely inaccurate predictions by March 2024.” The authors claim that by including a small fraction of 2024 data during training, the model continues performing accurately up to the present time. However, Figure 6 will suggest otherwise: the MLS tape recorder in 2025 differs considerably from the one estimated by OMPS LP, indicating that the model is already producing inaccurate predictions, presumably because the state of the stratosphere continues to evolve as the Hunga plume moves.

The rationale that “In 2025, the NNs perform reasonably well below 30 km, indicating that there is sufficient sensitivity for the determined approximation to remain accurate when applied to unseen data” may also be faulty, Hunga excess water vapor has not reached those levels, when it does, it may affect the estimates as well, as the training has not seen those type of values.

Overall, while the OMPS LP may accurately predict water vapor under “stable” conditions (i.e., prior to Hunga), it appears it cannot predict post-Hunga unless the training dataset includes some representation of those behaviors. Given that training with ACE-FTS and SAGE III/ISS failed, the absence of MLS data for training means the model will likely fail to capture the true state of the stratosphere as the plume evolves.

Furthermore, it is not clear whether the authors applied the MLS quality screening criteria to properly filter the MLS data. The MLS quality document is available at https://mls.jpl.nasa.gov/data/v5-0_data_quality_document.pdf

If the authors did apply the MLS quality screening, this should be clearly mentioned in the text, perhaps in the Data Curation section. If they did not, the analysis should be repeated using the quality screening to avoid retrieval artifacts, a priori influences, etc.

The same goes for ACE-FTS, the ACE-FTS screening criteria can be found at

https://doi.org/10.5194/amt-8-741-2015

Also, how does the ensemble standard deviation (used as an uncertainty estimate as discussed in line 121) compares with the MLS precision error. Are there comparable? What is the vertical resolution of the OMPS-LP water vapor dataset. Is it the same as MLS even though the OMPS-LP dataset will be reported in 1km spacing?

The manuscript needs to include a table discussing the OMPS LP water vapor characteristics (e.g., precision, horizontal resolution, vertical resolution)

Lastly, have the authors performed a feature importance analysis? In other words, how much of the water vapor information is coming from the LP radiances, how much from the FP-IT fields, and how much from the solar zenith angle? It is possible that the neural network primarily relies on temperature and pressure information to estimate water vapor, with the LP radiances contributing very little. Is it possible that some channels are contributing and other are not? How robust would the water vapor estimates be to jumps or changes in the reanalysis fields? The authors could modify these fields to assess their impact on the estimated water vapor.

Specific comments:

Given that the OMPS-LP water vapor dataset exhibits a different trend from the MLS dataset (Figure 8) and has a limited vertical range (11.5–40.5 km, though effectively only up to 30 km, as the manuscript states: “We therefore advise that users exercise caution when using the OMPS LP H₂O product above 30 km.”), the current title is somewhat misleading. A more appropriate title might be: “OMPS LP water vapor estimates based on a neural network trained on MLS water vapor.”

L2 the name of the instrument is: Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS) Please change accordingly here and elsewehere

L3 … Experiment III on the International Space Station *(SAGE III/ISS)*

L10 retrieve -> please change to estimate or predict, retrieve is associated with the typical retrieval process (i.e., optimal estimation).

L20 There is likely a better citation for UT water vapor, it has been known for decades. Please add other citations or at least add e.g., before Read et al 2022.

L28 for MLS please cite Waters et al 2006 https://ieeexplore.ieee.org/document/1624589

L29 For ACE-FTS please cite https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2005GL022386

L33 Please provide a brief description of the agreement between the instruments. These differences help set the stage for properly evaluating the OMPS LP water vapor measurements discussed here.

L35 now only operates *around* 6 days per month to *preserve measurement lifetime* and will continue …

L37 please add */ISS* after SAGEIII      (there was a SAGEIII meteop, so it is customary to call the SAGE III on the ISS, SAGE III/ISS).   Please change SAGEIII to SAGE III/ISS elsewhere as well.

L51 retrieve -> estimate

L60 eruption. You need to explain why you are using before and after Hunga. You never mentioned how Hunga altered the stratospheric water vapor.

L70 the same -> identical.    to ensure -> ensuring    

L71 differences -> variations

L73 How was the sensitivity to H2O estimated? Degrees of freedom, smaller precision, please provide succinctly the details.

L86 It is not clearly defined why the data needs to have aligned orbits, could the authors simply use all colocations that are within 6 hours and 100km. What is gained by using the 2 criteria listed in L82 and 83.

L88: Due this means that the training was based on around 8 percent of the available OMPS-LP data, i.e, 200000 colocations / (250000meas per year x10years). If it is, perhaps the authors should mention this on the text, to improve the context.

Figure 1 caption: Hunga peak in the lower stratosphere. So upper troposphere -> lower stratosphere

L89-90 where is this information coming from? All levels could be affected by apriori values, which is why the MLS data sets the precision to zero or negative. See MLS quality document.

The MLS quality document states for 316hpa “Occasionally erroneous low value < 1 ppmv and high value fliers are retrieved in the tropics, usually in clouds.”

L91-93 Are you using the closest FPIT fields to the MLS measurement locations or are you interpolating in time and space to the MLS measurements times and locations.  Please be specific.

L94 ACE -> ACE-FTS     (here and elsewhere)

L102 Are the FP-IT interpolated for the OMPS LP measurement times and locations, or are you using the closest fields in time and space?

L131 the correct citation for MLS v5 water vapor dataset is

Lambert, A., Read, W., & Livesey, N. (2020). MLS/Aura Level 2 water vapor (H2O) mixing ratio V005. [Dataset]. Goddard Earth Sciences Data and Information Services Center (GES DISC). https://doi.org/10.5067/Aura/MLS/DATA2508

L140 The drift was found in v4. In version v5 the drift was ameliorated. From Livesey et al (2021) “As a result of this correction, the MLS v5 H₂O record shows no statistically significant drifts compared to ACE-FTS. However, statistically significant drifts remain between MLS v5 and frost point measurements, although they are reduced.”

I think the authors can simply delete the mentioned of the drift, that is, “We investigate whether our product shows similar properties as the MLS product by …”

Or they can be more specific and say something like: Given the statistically significant remaining drifts between MLS v5 and frost point measurements, despite the applied drift correction (Livesey et al 2021),  we investigate …

L194 please add *upper* before tropospheric

L194-197. How was the presence of tropospheric clouds determined?  How was the presence of PSC determined? Through OMPS-LP measurements, MLS measurements, other? Please explain in the text. Could you show figures showing the lack of impact for tropospheric clouds as well as the impact due to PSCs?

L200. What was the coincident criteria for training ACE-FTS and SAGE III/ISS? How many coincidences were available. Why was SABER not considered for training?

L202  were negative -> were not satisfactory   or were suboptimal.

L209 on line 87 the authors mentioned 2 million colocations not 1 million, which one is the correct value?

L218 Why are you using the median and not the mean?

L220 Why are you using the standard error of the median? You should use the standard deviation as a metric.

L230-232. Please delete the following sentence “This behavior is generally consistent with earlier studies, such as Davis et al. (2021) which shows a similar pattern of increasing differences between 15–40 km when comparing SAGE and ACE.”   The fact that SAGE and ACE also have differences increasing with altitude is pure coincidence and have no merit in this discussion.

L235 our -> the

L264-267: This analysis is not enough to conclude that NN reduce these drifts. The authors need to collocate the MLS and OMPS data with the balloon measurements and compute the drifts for both datasets. Figure 8 only shows that the long-term trends are different between MLS and OMPS.

L275 How do the authors determine that OMPS-LP is more accurate than the natural variability of H2O? This could also mean that OMPS-LP is capturing less variability?

L278 Why is Figure 10 displaying SAGE II and SAGEIII (not ISS I presume). These datasets have not been discussed, I suggest deleting those lines.  Also, Figure 10 looks pixelated. Why was SABER not considered for this figure?

L307 should this be 30 km? after the tape recorder discussion.