The authors have done a thorough job addressing the reviewers’ concerns. That said, some issues were addressed only in the response to the reviewers and were not incorporated into the main text. Below, I identify the points that, in my view, should be integrated into the manuscript for it to be suitable for publication.

Also, consider a different layout for Figure 2, a figure / table is perhaps not the best way of showing this. I believe two bar charts (one for ascending and one for descending), displaying the number of sondes per day will be much easier to grasp (with one color for BCO, one for R/V meteor, and one for INMG)

Define GTS, SYNOP, TEMP, ECMWF, ERA6

Please make sure that all variables in the table 3 are explained, that is, what is ta, mr, dz, wspd, theta, q, etc. Although for many readers these variables are obvious, for others, they are not.


Points that need to be included in the main text:
RC1
2. To my knowledge, there have been many other Atlantic radiosonde launch campaigns, such as POLARSTERN, DBLK, HTXUH4H, among others. On an old hard drive, I discovered approximately 20,000 high-resolution radiosonde profiles launched over the ocean, most of which were conducted over the Atlantic. As an example, I have provided two screenshots of the data listing below. Although I have not systematically examined the spatial overlap between these data and the study area, it might be worthwhile to briefly introduce other oceanic radiosonde experiments in the introduction and provide a concise comparison. This would allow readers to gain a broader perspective on the full scope of Atlantic radiosonde campaigns.

Please include this sentence (or something similar in the main manuscript): A comprehensive overview paper of the ORCESTRA campaign is currently in preparation and will place ORCESTRA in the broader historical context of Atlantic radiosonde observations, including previous ship-based efforts.

5. The Introduction would benefit from a clearer explanation of the scientific motivation behind ORCESTRA. What were the key research questions or atmospheric processes that this campaign aimed to address?

There is nothing wrong with some repetition, at least include the ORCESTRA overall objectives as detailed in the ORCESTRA site (copy and pasted below):
Overall objectives of the ORCESTRA campaign
1. Determine the drivers of mesoscale organisation in the tropics and their impact on smallscale weather systems and the large scale circulation with a particular focus on the structure and variability of the Atlantic ITCZ
2. Serve as a benchmark for satellite remote sensing and a new generation of high resolution storm resolving models


RC2
1. The processing uses the vapor pressure equation of IAPWS (Wagner and Pruss, 2002). For radiosonde data, the vapor pressure equation used by the manufacturer should also be used in subsequent calculations to avoid any discrepancies between the calibration of the sensor and the measurements. At warm temperatures, this issue is irrelevant; however, at cold temperatures near the tropopause, this issue may become quite significant. Vaisala uses the equation by Hardy (1998) and Modem uses the equation by Sonntag (1994). Fortunately, the differences to the IAPWS formulation are small and reprocessing is not needed, but you should convince yourself of this difference and make a statement to that effect.

We compared several commonly used formulations of the saturation vapor pressure over
liquid water, including Wagner and Pruss (2002), Hardy (1998), Sonntag (1990, 1994), Murphy and Koop (2005), and others, against the IAPWS-97 reference formulation (c.f. Figure 1). The formulations by Wagner–Pruss, Hardy, Sonntag, and Murphy–Koop all exhibit very close agreement with the IAPWS-97 reference, with deviations typically well below 1%. In contrast, more commonly used Magnus- or Tetens-type approximations (e.g. Bolton, 1980) show substantially larger deviations.
Below the formal IAPWS-97 validity range, no definitive reference exists. The clustering
of deviations among simpler empirical formulations at low temperatures therefore does not imply a systematic bias in Wagner–Pruss, particularly given their inferior performance at higher temperatures and their limited physical basis. The closer agreement of Murphy–Koop with these formulations at low temperatures instead reflects its specific optimization for ice saturation.
We conclude that replacing the manufacturer-specific formulations (Hardy for Vaisala and
Sonntag for Meteomodem) with a single, consistent formulation does not introduce a meaningful bias in the derived humidity-related quantities for this dataset.

Please summarize this discussion into the main text (no need to include the figure, just summarize the rationale / results).

17. Line 199: Is the splitting into ascent and descent profile on geopotential altitude, or GPS altitude grid? There is a space missing after 10 m.
Ascent and descent profiles are separated using the radiosonde dropping flag, not by altitude. The profiles are subsequently binned on a 10 m geopotential-height grid.
Please mention this in the main text.

23. Line 271: Masking relative humidities > 100% prior to any binning or averaging can lead
to small negative biases.
The masking step was included as a precautionary consistency check. However, in the
dataset used for this study no relative humidity values exceeding 100% were found. Consequently, this step does not affect the binning or averaging.
Please mention this in the main text.


25. Line 315: Why do you exclude the soundings while the R/V Meteor is still in Mindelo?
After the arrival at Barbados, you showed that there is very little di!erence between the ship
borne and land-based soundings. In addition, it would be interesting to see a comparison between the Modem and Vaisala soundings at Cabo Verde similar as that you did in Barbados.
The soundings launched while the R/V Meteor was still in Mindelo are excluded only from
the IWV analysis in order to focus on open-ocean conditions. These launches occurred while the vessel was in port, were conducted sporadically rather than following the standard 8-per-day schedule, and were not accompanied by other shipborne measurements. Including them would introduce a small number of profiles influenced by coastal conditions, leading to a slightly drier tail in the IWV distribution that is not representative of the open Atlantic. The comparison between shipborne and land-based soundings at Barbados addressed a different question and was intended to assess platform effects rather than IWV characteristics. The R/V Meteor was stationed in coastal waters approximately 1.5 km offshore from the Barbados Cloud Observatory,
with the regular 8-per-day launch schedule and concurrent measurements maintained, enabling a meaningful comparison. A similar comparison in the eastern Atlantic was not pursued, as the vessel did not approach the INMG site with comparable spatial proximity.

Please mention / summarize this in the main text.

26. Line 326: For all sounding sites, how is the first line of the radiosonde profile handled? Is it a measurement from a nearby surface reference station, or the potentially biased radiosonde data, which is not yet properly ventilated? If an external surface station is used, then all profiles should have a surface value.
The first level of each raw (Level-0) radiosonde profile corresponds to surface observations
taken immediately prior to launch. For land-based sites, this first line is provided by collocated surface reference stations (e.g. BCO and the INMG station2 for MAESTRO), while for shipbased launches it corresponds to the onboard meteorological instrumentation. These surface values are therefore present in all profiles. An exception applies to two MAESTRO soundings launched from a slightly offset location (black crosses in Figure 1 inset b)), for which the first level corresponds to the radiosonde measurements at launch and may be subject to incomplete ventilation.
The Level-2 product used in the IWV computation here in line 326, is obtained by vertically
resampling the raw profiles onto a fixed altitude grid (0–31 km). Because the actual launch
altitude differs between platforms and sites (e.g. land stations versus ship deck), the lowest valid
measurement in the resampled Level-2 profile is not necessarily located at 0 m. As a result, the surface reference measurement present in the raw profile may appear at a higher altitude level in the Level-2 product, or may not coincide with the lowest grid level. Consequently, not all Level-2 profiles contain a valid value at 0 m. If near-surface humidity values are missing at the lowest grid levels, limited backfilling is applied to enable the computation of IWV.

Please mention summarize this in the main text.

27. Similarly, a temporary setup of a system is often prone to small configuration errors.
Did you compare the settings of the second sounding system on the ship with the permanently installed system? Are all heights correct, and are the launches detected properly without loss of data just after launch? You can verify this by comparing the raw and processed data in the mwx file.
On board R/V Meteor, the configuration of the temporarily installed sounding system
was cross-checked against the permanently installed system. Height measurement used for radiosonde initialization were taken from the same shipboard instrument in both systems. Data integrity immediately after launch was routinely monitored, with each launch observed for at least 10min to ensure stable telemetry, correct launch detection, and continuous data reception.

Please mention this in the main text.


28. Lines 357: Descending profiles are often not tracked to the surface, which makes it more difficult to get a good integrated water column. You could create Figure 10 using only ascent profiles. This would also reduce any hysteresis or other sensor issues that may exist on descent.
We agree that descending profiles are more prone to incomplete near-surface coverage and potential sensor hysteresis effects. For this reason, we applied the strict IWV filtering criteria described in Section 3.3 to ensure physically meaningful column integrals for both ascent and descent segments. However, we repeated the analysis using only ascent profiles. As shown by the comparison in Figure 3 in this document, the resulting IWV distributions remain qualitatively unchanged, with only marginal quantitative differences. The central tendencies and relative offsets between platforms are preserved. We therefore retain the combined ascent–descent dataset in order to maximize the use of available observations, while noting that the applied filtering effectively mitigates the issues raised for descent profiles.

Please summarize this into the main text (no need to include the figure, just mention that there is no difference when using only ascents or ascents and descents)

Thank you for addressing my concerns.