Paper Summary:

This paper analyses the random and bias error characteristics of data that was made available as part of the Radio Occultation Modeling Experiment (ROMEX). ROMEX is an agency and scientific community effort to quantify the benefits of assimilating increasing numbers of RO profiles for numerical weather prediction. In the three-month ROMEX experimental period, in excess of 35,000 profiles per day were made available to ROMEX participants who agreed to the terms and conditions of using ROMEX data. (This compares to less than 10,000 RO profiles per day that are available operationally.) The paper focuses on the three largest datasets from ROMEX: COSMIC-2 (C2), Spire, and Yunyao. The three-cornered-hat (3CH) method is used to determine the uncertainties of the C2, Spire, and Yunyao bending angles and refractivities. The analysis covers height dependence and geographic variations. Reanalyses are used as two corners of the 3CH method to estimate the uncertainty of the combined C2-Spire-Yunyao data set.

Review Summary:

The paper provides valuable scientific information on the ROMEX data, which is an unprecedented data set for RO and also involves new commercial sources of data such as Yunyao which has not been extensively analyzed in the past. This is the first time that such a large data set is evaluated for its error characteristics. Understanding the error characteristics is very important for data assimilation and numerical weather prediction. Before publication, the paper should clarify certain aspects of the analysis and address questions as detailed below. After suitably addressing these aspects, the paper should be ready for publication.

Citation of the literature is generally appropriate, but with reliance on an unpublished document that could be given a DOI. The reference Aparicio 2024 is to a presentation that could be put online with a DOI.

The material is presented well in a logical and clear manner.

Detailed Comments:

Line 107: a brief explanation of “excess phase data” should be provided so that the paper is accessible to a less specialized audience.

Lines 150-166: while the theory of the 3CH method is explained elsewhere, this brief summary does not fully serve the paper. The paper compares different data sets to one another, and no data set is claimed to be “truth”, so why is “truth” referred to here? Rather than truth, the authors might be referring to a reference data set for which biases and variances are determined relative to that reference. Wouldn’t these equations still hold if one of the data sets is viewed as “reference” instead of truth? Or are these equations only valid if one of the data sets is actually “truth”, which has zero bias and zero random error? The term “bias” in this paper appears to refer to a bias between two data sets, and not between one data set and truth. What is meant by “bias” in the paper should be clarified.

Another way this brief introduction is not serving the paper is that the paper analyzes various subsets of data for which it becomes clear there is not a single bias applicable to all subsets. For example, bias appears to vary geographically, and the authors apply the analysis to the global data set which is not characterized by a single bias. Is the global bias expected to be the mean of the regional biases? Is equation (1) valid for a dataset that is characterized by multiple biases? The authors should clarify how the equations 1-3 apply to the data sets being analyzed in the paper.

The following phrase is used on line 211: “but at the expense of fewer pairs in the sample and greater noise in the statistics.” The description in lines 150-161 does not contain terms corresponding to this “statistical noise”. For such a term to exist would require the concept of a sample mean as an estimate of the mean of a theoretical parent distribution. (Similarly for the parent standard deviation, etc.). The authors apparently are not concerned with statistical noise in their analyses, relying on large enough sample sizes to render the statistical noise negligible. The authors should make some reference to this implicit assumption in the paper.

Line 270: provide some sense of what “impact height” is and how it relates to geometric height, for the less specialized reader.

Line 276-279: please clarify this sentence. It is hard to understand.

Line 313-317: this paragraph and example should be better defined. There are numerous references that suggest RO can be used to advantage to improve predictions related to tropical cyclones (TC). What is meant by “resolve” here is not clear. RO inherently has relatively poor horizontal resolution per observation, which will not change with increasing numbers of observations. Even if RO cannot spatially “resolve” TC, increasing numbers of RO could improve predictions related to TC, such as intensification and track. This paragraph is not convincing regarding whether increasing numbers of RO have no benefit for TC.

Lines 324-325: if C2 does not exhibit the same count variation as Spire near the equator, it is worth commenting on why this might be case, if true. Both data sets sample the equatorial anomaly.

Line 452: please provide the quantity (approximate) of operational data assimilated so that this statement can be provided in better context.

Lines 459-460: weren’t the Yunyao data adjusted after the initial processing, so why does this artifact remain?

Line 502: Figure 7 is indeed impressive, but also somewhat puzzling. Whereas RO-RO comparisons have consistently shown growing uncertainties below 10 km, such uncertainty growth for the models is unexpected. For example, Figure 11 of Hersbach et al. (DOI 10.1002/qj.3803) for ERA5 temperature uncertainty does not appear to match what would occur with ERA5 bending angle uncertainty as indicated in Figure 7. While there is a modest increase of temperature uncertainty below ~8 km in the ERA5 paper, it does not increase dramtically towards the surface and is not much larger than uncertainty near 10 km. Please reconcile Figure 7 with Figure 11 of Hersbach et al.

Line 525: please clarify what BFRPRF in Figure 8 refers to. What are the dashed blue and green lines?

Line 531: processing provenance is somewhat unclear. Were the Yunyao data used here reprocessed by EUMETSAT or UCAR?

Line 583: please clarify what is the location of the ERA5 model and how is that determined. Is it a nearby grid point value? Isn’t it straightforward to interpolate ERA5 values on a grid to an RO location, thus eliminating collocation error for all ERA5 comparisons?

Line 594: what is meant by short-range forecast of a reanalysis and why use that rather than the reanalysis value that is based on all contemporaneous data?

Line 620: how is it known that above 30 km ERA5 biases are dominant?

Line 627: same comment – how is it known that ERA5 biases are dominant?

Line 718: missing close of parentheses.

Line 747: this suggests that the bias caused by Rc should not create model biases after assimilation because N does not exhibit the bias. The forward operator applied to bending angle should remove the bias. Please reconcile the small N bias with a bias in the model that assimilates BA. Note that Zhou et al. (DOI 10.1029/2024JD041295) detected tempature biases in C2 data. How do such biases arise if N is not biased due to Rc variations?

Lines 760-773: the DD examples used here relate to reducing collocation error. It’s not immediately clear how DD reduces Rc error. Please provide more details regarding the algorithm that reduces Rc error. Include a discussion of how bending angle is computed from the model.

Lines 800-802: so the claim is being made here that the local radius of curvature computed for each occultation is only computed at one altitude, and not repeatedly as the raypath drifts?

Line 808: What of negative azimuths?

Line 848: this section should explicitly note whether bending angles or refractivity are being referred to.

Short Summary:

The paper analyses data from the ROMEX experiment focusing on the three missions that contributed most data in the ROMEX period (September - November, 2022). These are Spire, Yunyao, and COSMIC-2. Uncertainties and biases between datasets are addressed and sought understood. There are many interesting results, in particular global maps of uncertainties and biases, that I think deserve publication. However, I have concerns that the data that are analyzed are different from the data that have been provided to NWP users in the ROMEX experiment (if I understand it correctly). And I'm unsure if the data that are analysed are publicly available (or maybe not yet). I also have concerns about the correctness of some of the statements, which I elaborate in my specific comments below.

In my opinion, the paper needs major revision, with an update of some of the figures and a few more added, as well as modifications to many of the statements in the paper. Also more references needs to be included. The description of averaging method needs improvement. Generally the paper is well organized and the study is important.

Specific comments:

L26: Maybe the COSMIC acronym should be spelt out. Or maybe it could be handled by saying "... referred to as COSMIC-2 (C2) ...".

L33-34: "They are similar on the average in the region of overlap (45°S-45°N)". I don't think you can say this. According to Fig. 8, the Yunyao uncertainties are significantly larger (up to a factor of two) than Spire and C2 in the 10-15 km range. Is the sentence needed?

L37: "The assimilation of ROMEX data caused small degradations in biases in several NWP models". I suggest to remove this sentence from the abstract since it is not part of the study, and it is not mentioned anywhere else in the paper. It may give the impression that the biases studied in this paper was affecting the assimilation experiments, but I think it is now understood that they are likely not the cause of what NWP centers have found (presented at ROMEX workshops). I also question whether the data analysed in this study are similar enough to the data that were assimilated (see later comments).

L43-45: Maybe this sentence in parenthesis is not needed in the abstract. It is explained later in the paper, and the detail is not needed here.

L44: BA is not defined (but see my suggestion above of removing the sentence).

L68-69: "The ROMEX data became available at the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) ...". I think you should add here "Radio Occultation Meteorology (ROM) Satellite Application Facility (SAF)". Then in line 123: "EUMETSAT Satellite Application Facility (SAF) on Radio Occultation Meteorology (ROM)" can become EUMETSAT ROM SAF.

L87-88: "Spire and COSMIC-2 (C2) are relatively well known and have been widely studied". Please provide references.

L91: "... Germany 17-19 April 2024 (Cheng 2025)". Should it be "(Cheng 2024)"?

L106-107: "The original (raw) data were downloaded ... by each data provider". Could you clarify which data providers you mean? It is not clear to me why data providers would need to download their own data.

L113-118: When was this done? Does this cover all ROMEX data or only a subset of them? Did EUMETSAT process these into refractivity?  It appears that these data are different from the data described in the next paragraph (line 122 and forward), but it should be made much more clear up front why you mention these data and if it is those that are analysed in this study (or maybe it is only the UCAR part of them - needs to be more clear).

L116-118: "EUMETSAT, UCAR, and STAR processed the excess phase data into bending angles, refractivities, and other products, as described by Kuo et al. (2004) and Steiner et al. (2020)". But these references did not describe the processing of ROMEX data. Text needs modification to be clear. Perhaps mentioning EUMETSAT and NOAA STAR here just adds confusion.

L130-135: I think you should move the citation in line 135 to line 130 as "Aparicio (2024)" instead of saying "Josep Aparicio (Environment Canada)". It is a problem that this work is not publicly available. Did you check with Josep Aparicio if a publication is forthcoming?

L138-141: "... data that were originally provided to EUMETSAT and then processed by UCAR ...". Is this the data mentioned earlier (I think so, but it is not clear)? Are these data available anywhere (I could not find them)? In line 140 you say that UCAR-processed data and EUMETSAT-processed data are similar, but couldn't there be important differences larger than the differences between some missions? Three potential examples comes to mind after reading the whole paper: 

1) Is there a different penetration for Metop in Fig. 4. than what was provided to NWP centres? 

2) What is the influence of different definitions of the occultation point? 

3) Could refractivity biases be significantly different at high altitudes due to different approaches in statistical optimization?

L139: "In sensitivity studies to investigate structural uncertainty (Steiner et al. 2020), we find ...". I think you should skip the citation as it is not directly related to the data analysed here. Or phrase things differently.

L141: "... processed data are similar. Examples are shown in the Supplement.". Please provide the figure numbers?

L144-146: "The penetration rate is defined as the percentage of successful occultations reaching different levels ...". Is it one number? or several numbers? It is not clearly defined here or anywhere else. Maybe it should not be called "rate".

L160-166: "The root mean square error is ... RMS" Why mention it if it isn't used?

L172: "RO biases are therefore assimilated in NWP models without bias corrections". Do you mean "RO observations are therefore assimilated ..."?

L180-183: "... commonly used verification charts ... often plotted together on a scale of -20% to +20%". Please provide references.

L184-189: Please provide specific references to support the statements here (I suppose from the ROMEX workshops).

L194: "sometimes called the RO core region, golden zone, or sweet spot" Please provide references.

L200-204: It would be good to mention here that the analyses and forecast data are forward modelled to BA and N.

L210: "Nielson" should be "Nielsen".

L269: "We compare bias and uncertainty profiles...". I suggest to say instead: "We compare bending angle observations ...". It is only the BA observations that are compared on impact heights, not the refractivity.

L269-272: Would it be more correct to say that the BA observations depend on five parameters: time, impact height, latitude, longitude, and azimuth? The dependency on azimuth is ignored in the comparisons, but not the dependency on the other four parameters.

L276: "The issues ... are much smaller than 0.15% ..." Issues are not quantitative (the representativeness differences are). Please modify text.

L278: "comparing RO missions on impact height ...". I suggest: "comparing RO missions ignoring the azimuth dependency ..."

L280: "However, they likely ...". Now 'they' refer only to issues, not to the representativeness differences. Please modify text.

L321: ... "in the 40-45°NS bands". I suggest to note here what NS means.

Figure 1: The first five figures span four pages, but are numbered 1a to 1e, with separate captions and without any direct relation. I suggest to number them 1 to 5. Similar could be said about Figure 10a and 10b,c. (They could be two separate figures).

Figure 1a: It could be mentioned in the figure caption that the underlying world map is for 12 UTC, but that it is not important. The x-axes could be labelled "local time [hr]".

Figure 1e: This plot gives the impression that there are very few occultations at high latitudes for the polar missions. It would be more interesting to see the density per unit area (divide by cos(lat)), which I think would be more relevant.

Figures in general: There is often information in the titles above figures which should be in the captions, e.g., "20 km" in Fig. 2 should probably be mentioned in the figure caption. Please check all figures and modify captions as needed.

L374: "Fig. 2 shows the daily BA profile counts (after UCAR CDAAC QC but before the 3CH QC)". What is the 3CH QC?

L433: "penetration rates are noticeably less for Metop-C (green), ...". And Metop-B too, it seems (I would say that the colors for Metop-B and Metop-C are olive green and/or dark cyan, but maybe there are more precise color names provided by the plotting sofware that could be used here). Why is the penetration so poor for Metop? Is it the same in the ROMEX core data that can be found at the ROMEX data server, or is this an artefact of the CDAAC processing? 

L436: "These results confirm that radio occultation is a useful method of obtaining global information on the planetary boundary layer." Please provide references (or maybe the sentence is not needed here).

L448: "In this section we present an overview of the bias and uncertainty statistics of all the ROMEX data. Many additional detailed results are presented in the Supplement.". But the supplement only contains details about the three missions that are the focus in the next sections, not the rest of the ROMEX data that are discussed in this section (Section 3). Maybe the last sentence is not needed.

L451-455: Say "MSL altitude" instead of "MSL height" (altitude is commonly used with reference to MSL). It would be good to mention here that it is the MSL altitude of the tangent points (I suppose).

L457: "... having the smallest uncertainties because of their more accurate clocks (Padovan et al. 2024).". Padovan et al. only looked at Sentinel-6 and COSMIC-2 and did not compare to other missions. Could the CDAAC processing also here have an influence?

Figure 5: Are all the missions processed by CDAAC with the same software version? If not, I suggest to add a table which shows processing versions and/or origin of the processing (e.g., if the low level processing was done by another data provider).

L458: "Fengyun-3 shows higher uncertainties between 10-30 km than the other missions. Yunyao has a peak in uncertainties between 10-15 km ...".  Should it be "uncertainty" (singular)? I don't think it is correct to say "between X-Y km". Either say between X and Y km, or in the X-Y km range (or similar). Please modify all such occurrences throughout the paper.

L468: "... ROMEX missions between 10-35 km (Fig. 6a)". This is the first reference to Fig. 6, but before that, it should be properly introduced in the text (not in a parenthesis). I noticed the same regarding Fig. 15 and Fig. 17.

L475: "Fig. 6: Upper left (a) ...". Letters (a,b,c,d) are missing in the Figure. Please add letters to each of the four panels.

Figure 6: There is a confusing background grid with dots over all four panels that doesn't belong here. It needs to be removed. Same can be said of Fig. 4.

L486-487: Perhaps replace "in these latitudes" with "at these latitudes".

L489: "figures show the importance of comparing different RO missions using spatial and temporal sampling as similar as possible". Why temporal? At first glance, I don't see any effect of different temporal samplings. Please clarify.

L503: "... the simple but effective error model used by ECMWF (Ruston and Healy 2022) is shown as a dashed line." I assume you mean "error model for RO observations". The general reader could misunderstand this to be the error of the ECMWF operational model, especially when one looks at the legend in Figure 7. I suggest to use a more precise indication than "ECMWF" in the legend to distinguish the meaning from that of "ERA5" and "JRA3Q".

L504: "Considering that it was developed many years ago, the agreement with the CSY data is remarkable.". I think this is an overstatement. The ECMWF-used error estimate is significantly larger at high altitudes. Please modify text.

L533: "uncertainties and biases below 5 km are related to their cutoff of carrier phase data ..." This is the first mention of a bias in Yunyao data below 5 km, and it is not shown in Fig. 8. Should it be shown?

L534: "... too early, as described by Sokolovskiy (2014) and noted by Marquardt et al. (2024).". Sokolovskiy (2014) did not describe Yunyao data, so I don't think you should cite it like this. In the conclusion of (Xu et al. 2025) it says: Larger biases are primarily observed in the lower troposphere, a phenomenon that has been extensively discussed in previous studies (Sokolovskiy et al., 2014; Xie et al., 2010). However, I think that is with reference to the more recent data version that Yunyao has processed. Are you sure that the much larger bias in the Yunyao data in your study can be attributed to the effects described by Sokolovskiy (2014)?

L535: I think it should it be (Xu et al. 2025).

Figure 8: It is very difficult to distinguish solid, dash-dotted, and dashed of the same color. Given the curves in Fig. 7, would it make sense to show differences to these in Fig. 8? Like Fig. 6, this figure also has a confusing background grid that needs to be removed.

L551: "Larger uncertainties occur over Asia and the Pacific". I see it in many other places in the tropics. Why single out Asia and the Pacific? It is difficult to see that the uncertainty is particularly larger there.

L552-554: "... interesting regional feature is the maximum uncertainty over the Weddell Sea at 20 and 30 km, which may be related to the ionospheric Weddell Sea anomaly (Chang et al. 2015).". If it is related to ionospheric disturbance, then why would it not be seen at 50 km? Could there be other explanations? It is indeed very interesting!

L564: "Fig. 9: Global distribution of 3CH uncertainties (%) for CSY BA at 3 km (a), 5 km (b), ...". Letters are missing in the figure panels.

L570: "Larger versions of the panels are presented in the Supplement." Actually, the Fig. 9 and Fig. 12 maps are a bit more sharp (when zooming in) in the paper than in the Supplement in my pdf viewer (acrobat).

L574-576: "The small negative impact of the ROMEX data on the biases of several NWP models has caused intensive study of possible causes of these small biases, including the possibility of small biases in the ROMEX data." Would a reference to 'this issue' (or similar) be possible here? (I assume that the paper, when it is in its final form, will be included in the special ROMEX issue).

L584-585: "... collocate an RO dataset with a model or another RO dataset, with advantages and limitations discussed in Section 2.". Could you briefly remind the reader what the advantages and limitations are? It is not totally clear to me.

L586-587: "... the two RO datasets in latitude-longitude bins and compute the averages over each bin of the difference between the RO and reference data (e.g., ERA5).". This raises a number of questions:

1) How do you handle drifting tangent points? Can one part (say the lower part) be averaged within one bin, and another part (say the upper part) be averaged within another bin? Or is it at a reference location for the whole profile? 

2) Is area weighting with a cosine factor taken into account when averaging over latitude? 

3) Are sampling errors taken into account? 

You need to describe your method in much more detail, possibly with equations, so that others can reproduce it. Gleisner et al. (2020) (https://doi.org/10.5194/amt-13-3081-2020) gives a nice description of how one can average in 5 degree latitude bins, including sampling error correction. I suppose your method is slightly different. Please describe it.

L587: "This results in large samples and all RO data can be included.". Do you mean all three months? Please clarify.

L602: "propagates into N after the Abel inversion". Since this is the first mention of the Abel inversion, perhaps a reference to the Abel transform would be good here for the general reader.

L617: "computed from 5x5° latitude-longitude bin averages over all bins and days of ROMEX." Do small bins at high latitudes contribute with the same weight as large bins at low latitudes?

L618: "bases" should be "biases".

L630-631: "Fig. 11 shows the mean differences between Yunyao and C2 from Spire between 10-40 km impact height.". I suggest rephrasing to something like: "Figure 11 shows Yunyao and C2 normalized biases relative to Spire between 10 and 40 km.".

L634-637: "The bulge between 15 and 20 km is likely related to the relatively large horizontal sampling differences in the 5x5° latitude-longitude bins in a layer with large variations of atmospheric densities in the vicinity of the tropopause." I don't understand how that could be a reason. Please elaborate and please clearly describe how the results in Fig. 11 are obtained. Do they come from collocations? What does "2-day Mean" in the title above plots mean?

L651-652: "ERA5 biases may be of comparable or larger magnitude at all levels." I don't think ERA5 biases are this large at the lower levels. Would you agree?

L670-672: "The ROM SAF Matched Occultation page presents daily estimates of the biases of RO satellites compared to other RO satellites, with a collocation criteria of 300 km and 3 hours (https://rom-saf.eumetsat.int/monitoring/index.php )." I think the collocations are more precisely at https://rom-saf.eumetsat.int/monitoring/matched.php.

L672-674: "This monitoring site shows various combinations of mean and standard deviation of differences between BA and N from different satellites". I don't think the sentence makes sense (Combinations of mean and standard deviation? Combinations of differences between BA and N?). Please rephrase.

L674-676: Perhaps the paper should not be discussing details in a figure from someone else's website. Is it necessary?

L694: "Why are C2 BA positively biased to Spire ...". Maybe say "biased relative to".

L711: "computed from 5x5° latitude-longitude bin averages over all bins ...". I think there may be a copy-paste mistake here. There are no averages over bins in Fig. 14, right? 

L716: "The small positive biases of C2 relative to Spire and other ROMEX missions in BA". Suggestion: "The small positive BA biases of C2 relative to Spire and other ROMEX missions".

L717: "... between 10 and 35 km result from their different orbits around the non-spherical Earth". I suggest to say "result from their different orbit configurations.".

L719: "... that best fits the Earth’s surface curvature at a given location ...". Ending parenthesis is missing. I would rather say: "... that fits the Earth's surface curvature along the direction of signal propagation ...".

L722-723: "The effect of this variation of Rc on the BA ...". Perhaps say "the azimuth" instead of "this" to be clear.

L723: "... may be called the anisotropy of Earth’s curvature effect.". Do I understand correctly that this covers both the azimuth effect and the sideways sliding effect? Should it be plural (effects)?

L724-725: "C2 is in a low-inclination orbit (24°), with all of its observations located within ±45° latitude and occultation planes predominantly oriented in an east-west (E-W) direction.". I'm not convinced of this. I suggest to include a figure showing the distribution of occultation plane azimuth angles for C2 and Spire.

L741: "... oriented E-W (as in most C2 occultations) ...". Please show it.

L746: "larger Rc will accumulate a slightly larger bending angle, due to traversing a slightly longer path." My understanding is different: A signal is being affected by slightly more perpendicular gradients (due to the slightly larger curvature) on its path in the vicinity of the tangent point (vertical refractivity gradients are precisely perpendicular at the tangent point, but gradually less perpendicular away from the tangent point because of the Earths curvature), thus accumulating more bending. The path length may in principle be the same. Would you agree with this?

L747-749: "Although this effect is small, it can still cause a difference up to about 0.3% in the bending angles measured at the same impact height along the equator between the N-S and E-W directions". Where does the number 0.3% come from? How did you calculate it?

L748: Perhaps "at" instead of "along".

L750: Better to say "as a function of impact parameter" instead of "as the function of the impact parameter".

L760: "The azimuth angle effect can be corrected ...". I suppose you mean when comparing collocated BAs on impact heights. Please make that clear in the text.

L760-761: "... through double differencing (DD) using a model." I suggest to call it sampling error correction, not DD. Schreiner et al.  (2019) used the term "sampling correction". The term DD in RO has been used for something else (to eliminate clock errors in the phase data by simultaneously observing two GPS satellites and using ground station data).

L784-785: "In RO data retrieval, it is commonly assumed that the occultation plane remains fixed throughout an occultation event and is anchored at the assigned occultation point, ...". My understanding is different: The centre of curvature and radius of curvature remains fixed, but the occultation plane is not assumed to be fixed. For each sample during an occultation a new occultation plane is calculated based on the fixed centre of curvature and the varying satellite positions. Thus, the occultation plane is assumed to change during the occultation, and the tangent point locations are different for each sample. We say that the tangent points are drifting. What is commonly assumed is that the surface of the Earth locally coincides with the surface of the sphere whose origin is at the centre of curvature. That is the assumption which does not hold exactly. It only holds in the direction of signal propagation at one instant during the occultation, the one that defines the occultation point which is then fixed. It doesn't generally hold in any other direction or at another instant. Would you agree with this?

L801: "can be corrected by adjusting the impact heights ...". Add "and altitudes". 

L801-802: "... by a correction factor termed the sideways sliding correction." Use of the word "factor" implies that something should be multiplied by a factor, which is not the case. Text needs slight modification.

L808: "Fig. 17: Difference in radius of curvature (dRc in km) along and across ray path ...". I suppose it is "across" minus "along" that is displayed in the figure, but the text is not clear on that. Please clarify what is subtracted from what.

L812-815: "Tests of the impact of the sideways sliding correction by EUMETSAT and UCAR demonstrated that the vertical variation of the effect depends on how the nominal location or point of an occultation (termed occultation point by UCAR and georeferencing by EUMETSAT) is defined ...". Is there a reference that could be included here?

L827-831: "The sideways sliding correction results in a small reduction in the average positive C2-ERA5 BA and N biases in the UCAR-processed data by approximately 0.05% in the stratosphere. It also corrects negative biases associated with polar orbiting satellites, mostly in the tropics, by a similar amount. At higher latitudes, the effect on observations from polar orbiters is negligible (less than 0.01%).". Please provide references.

L834-835: "... and the resulting C2-Spire bias is shown in Fig. 18. The reduction is smallest at 10 km ...". Lower than that, I suppose (but not shown). What is the typical altitude (or range of altitudes) of the tangent point for which the excess phase is 500 m?

L835: "... because of the definition of occultation point in the UCAR data". Insert "the" in front of "occultation point".

L839: "Fig. 18: Bias of C2 relative to Spire for UCAR standard (solid profiles) ...". Perhaps skip the word "profiles" here.

Figure 18: I suggest adding a plot showing the same for refractivity.

L845: "For C2, the off-boresight angles exceeding 40° are mostly concentrated between 40-45°N/S". It would be interesting to see this in a figure.

L852-857: "An average of 35,000 RO profiles per day from 13 different RO missions from the United States, Europe, and China are being used in NWP models at major international centers to study how different numbers of RO profiles affect the analyses and forecasts. This paper evaluates the characteristics of the ROMEX data used in these experiments, with emphasis on the three largest datasets, COSMIC-2, Spire, and Yunyao." My understanding is that the data analyzed in this paper are not the same as the data provided to the NWP centers for all missions (C2 may be an exception). There may be differences in the biases and uncertainties due to different processing, and I have pointed to some examples in my comments above (L138-141). It is fine that this paper analyses the UCAR-processed data, but I don't think you can write that it is generally the ROMEX data that are being used in experiments at NWP centres. Please modify text.

L884-885: "This apparent bias is investigated and found to be a result of their different orbits.". I think the statement is correct that it is due to different orbits (or orbit configurations). However, I don't see any real evidence in the paper. I see a reduction in C2 bending angle biases when doing DD (sampling error correction) in Fig. 16, but that was already shown by Schreiner et al. (2019). Could more evidence that the reason is to be found in the different orbit configurations/azimuths be provided (cf. my comments to L724-725 and L747-749)?

L889: "... is different azimuth or viewing angles on the average ...".  This is the first time "viewing angles" are mentioned, and they are not discussed further. Please clarify.

L895: "... sideways sliding of the occultation planes during occultations.". I suggest to say "... sideways sliding of tangent points" (cf. comment to L784-785).

L896: "This sliding results in different radii of curvature of Earth ...". The sliding does not result in different radii of curvature, they are already there. Please rephrase.

L895-896: "... and different impact parameters ...". The impact parameters are not changed, only the height of tangent points (which are more accurately referenced to the surface of the ellipsoid instead of to the surface of a reference sphere). Please rephrase.

L897-899: "... and creates a positive bias of about 0.05% in the UCAR-processed C2 bending angle (BA) and refractivity (N) observations in the stratosphere compared to those of the polar orbiters". The positive bias seems less than 0.05% in Fig. 18. (perhaps half of that at 30 km where the difference between C2 and Spire is largest).

L899: "The sideways sliding effect was identified and discussed by Josep Aparicio in November 2024. It can be easily corrected in the processing of the RO data by applying a correction to the impact height.". I don't agree with this statement. My understanding is this: The impact height is not involved in the processing of RO data (the impact parameter is involved, but it is not changed). The impact height is used when we compare bending angles on a vertical scale (where we then subtract the radius of curvature from the impact parameter), and it is here we need to make a correction. Would you agree with this?

L906-907: "Code and data availability. The ROMEX data are available free of charge through ROM SAF under the ROMEX terms and conditions.". Please give precise information which ROMEX data were used. Are you sure the data processed by UCAR and STAR are on the ROM SAF server? I was not able to find it. There are different datasets in different versions. Which ones were used for this study?

L918: "JS assisted RA and JS by ...". There are two JS's.

L980: I think "Chen" should be "Cheng".

L985: "Presentation at the 1st ROMEX Workshop April 17, 2024 at EUMETSAT, ..." I think you mean: "Presentation at the 2nd ROMEX Workshop February 27, 2025 at EUMETSAT, ..."

L1077-1081: The reference can be updated since the final revised paper has been published.

Supplement:

There are detailed discussions in the figure captions of Figs.  S1.2-S1.8, which corresponds to Fig. 12 in the paper. Wouldn't it be better to have these discussions in the text in the paper?

More information on what is shown in the three panels of Fig. S2.1 should be added to the figure caption.

I think some of the N bias maps (Figs. S2.7 and S2.8) depend significantly on the use of a climatology in the statistical optimization at high altitudes (cf. my comment to L138-141.).

Fig. S3.1: Why show this figure again?

Fig. S6.1: It seems that N is affected by climatology above 30 km, and there is no positive bias relative to ERA5 at 40 km, as there is in bending angle (Fig. S5.1). Could you comment on it in the Supplement (or in the paper)?

Fig. S6.6: "All three missions have similar distributions.". I think that is an overstatement. I see significantly larger positive biases for C2, which is also seen in Fig. S6.1. Do you have an explanation for that? (cf. also my comment to Figure 18 in the paper).