Thank you for presenting this new GNSS dataset, which holds potential for climate studies. I found the manuscript well-structured and written, and the methodology sound. However, I would like to share a few comments and questions regarding some of the methodological choices and propose suggestions to further enhance the QC/QA process and overall quality of your dataset.

1) Section 2.2 GNSS data processing: While you mention adhering to the highest standards in your study, it is worth noting that the analysis was conducted using Bernese GNSS Software version 5.2. Since 2022, version 5.4 has been available, introducing several improvements. These include enhanced observation (RINEX) quality control and preprocessing, improved ambiguity resolution for PPP, and updated tropospheric models such as VMF3. Considering these advancements might further strengthen the robustness and quality of your dataset.

2) Section 2.3 Retrieval of PWV: the calculation of ZHD from the numerical integration of refractivity is not recommended (Jones et al., 2020, chap. 5.4.2), especially when only 37 pressure levels are available such as with ERA5. Instead, Saastamoinen formula should be used with surface pressure which can still be computed from ERA5 (with adequate interpolation between levels). This is the approach actually used by Haase et al. (2003) for their GNSS ZTD to IWV conversion. Note that they use only Eq. (2) for the integration of radiosonde profiles, which have many more vertical levels.

3) Section 3.2 Screening based on GNSS-ZTD results only: proper credit should be given to Bock, 2020, who introduced the general approach and the methodology to choose range-check and outlier check limits for GNSS ZTD and formal errors which are actually followed here.

To clarify the purpose and usage of the formal errors for the screening process, it may be judicious to more section 4.1 here.

4) Line 225 to 260: referring to systematic biases may be misleading here. Referring to the “consistency of ZTD estimated from collocated GNSS sites” as you do later on (Line 256) seems more precise.

I have one concern here with the impact of height differences. As you mention on line 238, “discrepancies … are often attributed to height differences”. This effect is actually expected. A simple rule of thumb approach predicts bias in ZHD around 10 mm and a few mm from ZWD as well based on a 50 m height difference. Such biases can be avoided by applying a proper vertical correction such as described in Bock et al., 2022. Following this approach can be very valuable in detecting station-specific biases when several nearby stations are available.

Another concern is with the impact of equipment changes which can mask short-term site-specific biases when computed over long periods. This should mentioned here and also underlines the importance of offset detection that is discussed later.

Line 235 to 250: the discussion employs the terms “bias, deviations and differences”, are these referring to mean values? Please clarify. Separating the mean and standard deviation of differences rather than using the RMS (which mixes both) would also give more insight into the nature of the differences.

Line 260: “additional quality control” would better fit here in place of “additional screening”.

5) Section 3.3 Screening based on comparison with reference PWV data: eliminating the negative IWV values may not be sufficient and may actually be hiding a more general bias between GNSS and ERA5 (possibly with seasonal variation). To avoid this caveat, it is preferable to compare ZTD values (as also recommended in Jones et al., 2020, chap. 5.4.1). Then you would probably notice a bias and decide to remove the flawed stations or find that representativeness errors in ERA5 are the limitation.

6) Line 274-292: I don’t understand the rationale of comparing GNSS PWV at a target station to ERA5 PWV at nearby stations. Here you mix to types of errors: GNSS vs. ERA5 (different data sources) and target vs. nearby (difference due to distance). If this procedure is inspired by Nguyen et al., 2024, it is actually not relevant here. Please clarify or correct.

Figure 8: this PWV difference series is really suspect. It looks like the ERA5 values in your GNSS – ERA5 differences are very close to zero. Please check.

7) Section 4.1 Formal errors in ZTD estimations: consider moving this Section to Section 3.2.

Give the % of the CDF corresponding to a formal error of 10 mm, which is the limit used for the range check in Section 3.2.

Line 299: “The formal errors of the estimated ZTD are known to play a key role in analysing the quality of GNSS”. Although formal errors may help in the QC/QA of GNSS ZTD estimates (Bock, 2020), it is an overstatement to say that they play a key role. Please moderate.

8) Section 4.2 Cross-comparison of PWV with external references

Line 232: the description of ERA5 (number of pressure levels, horizontal resolution, etc.) should be given earlier, e.g. in Section 2.3 when ERA5 is first introduced and used.

Line 383 and 405 + Line 229 (Section 3.2): explain why you chose three different collocation limits for these comparisons (GNSS vs GNSS, GNSS vs. VLBI, and GNSS vs. RS).

9) Units for ZTD and PWV comparisons: use mm for ZTD and kg/m2 for PWV to avoid confusion.

I am wondering whether some of the published GNSS vs. VLBI comparisons are not cited in the wrong unit (e.g. large biases and RMS values).

10) Section 4.3 Offset detection: my main concern here is that you applied the detection to a subset of stations only (less than 50% of all your stations). Considering the use of this dataset for climate studies (e.g. trend analysis), this is a serious drawback. Please explain why not all stations have been checked and whether you intend to further complete this.

Another couple of questions concern the confidence that can be placed in the PMTred method and the validation of the detected change-points. First, the number of change points with this method seems a little underestimated compared to studies using other methods (Venema et al., 2012; Van Malderen et al., 2020; Nguyen et al., 2021). Second, the validation with GNSS metadata is not robust as some equipment changes may be missing and not only equipment changes are suspected to produce offsets, but also environment changes. In addition, the numerous firmware changes are not expected to have a significant impact and instead lead to accepting many false detections. To overcome these limitations, it may be advisable to cross-compare the results from different detection methods and to implement a more robust validation method, e.g. based on multiple pairwise comparisons (Caussinus and Mestre, 2004; Menne and Williams, 2009; Nguyen et al., 2024).

Regarding alternative methods, Van Malderen et al., 2020, evaluated several of them in the similar context as the present study. Their benchmark study showed that PMTred, which is based on tests, is not performing well with the type of data used here (GNSS minus reanalysis differences). The best methods are indeed based on penalized maximum likelihood. One of them was first published under the name GNSSseg by Quarello et al., 2022, and recently updated and renamed PMLseg by Nguyen et al., 2025. This method uses penalized maximum likelihood and was specially developed for the segmentation of GNSS minus reanalysis differences. The authors may find it interesting as an alternative or simply for cross-checking their PMTred results.

11) Figures 20 and 21: the interpolation of 2D fields over the ocean looks very unrealistic. It may be preferable to mask the oceans in these figures.

Line 596: I was wondering if ENSO was not a more likely explanation for the interannual variability observed in these figures.

12) Section 7: Summary and Outlook: the text of this section seems to overstate a little the quality of the dataset given the mentioned limitations.

While the dataset could benefit from further homogenization and the use of the latest GNSS processing software, this work represents a valuable contribution that will enrich discussions and collaborations within the IAG community (e.g. within the ICCC joint working group C.8 on Optimal processing and homogenization of GNSS-PW climate data records).

References

Bock, O. (2020) Standardization of ZTD screening and IWV conversion, in: Advanced GNSS Tropospheric Products for Monitoring SevereWeather Events and Climate: COST Action ES1206 Final Action Dissemination Report, edited by Jones, J., Guerova, G., Douša, J., Dick, G., de Haan, S., Pottiaux, E., Bock, O., Pacione, R., and van Malderen, R., chap. 5, pp. 314–324, Springer International Publishing, https://doi.org/10.1007/978-3-030-13901-8_5.

Bock, O., Bosser, P., and Mears, C. (2022) An improved vertical correction method for the inter-comparison and inter-validation of integrated water vapour measurements, Atmos. Meas. Tech., 15, 5643–5665, https://doi.org/10.5194/amt-15-5643-2022

Caussinus, H. & Mestre, O. (2004) Detection and correction of artificial shifts in climate series. Journal of the Royal Statistical Society Series C: Applied Statistics, 53(3), 405–425. Available from: https://doi.org/10.1111/j.1467-9876.2004.05155.x

Jones, J., Guerova, G., Douša, J., Dick, G., de Haan, S., Pottiaux, E., Bock, O., Pacione, R., van Malderen, R. (Eds.) ; Advanced GNSS Tropospheric Products for Monitoring Severe Weather Events and Climate: COST Action ES1206 Final Action Dissemination Report (2020), Springer International Publishing: Cham, Switzerland, 2020 ; https://doi.org:10.1007/978-3-030-13901-8.

Menne, M.J. & Williams, C.N. (2009) Homogenization of temperature series via pairwise comparisons. Journal of Climate, 22(7), 1700–1717. Available from: https://doi.org/10.1175/2008jcli2263.1.

Nguyen, K. N., Bock, O., & Lebarbier, E. (2024). A statistical method for the attribution of change-points in segmented Integrated Water Vapor difference time series. International Journal of Climatology, 44(6), 2069–2086. https://doi.org/10.1002/joc.8441

Nguyen, K. N., Bock, O., & Lebarbier, E. (2025), PMLseg: a R package for the segmentation of univariate time series based on Penalized Maximum Likelihood. https://github.com/khanhninhnguyen/PMLSeg/tree/main

Quarello, A., Bock, O., Lebarbier (2022) GNSSseg, a statistical method for the segmentation of daily GNSS IWV time series, Remote Sensing, 14, 3379, https://doi.org/10.3390/rs14143379

Van Malderen, R., E. Pottiaux, A. Klos, P. Domonkos, M. Elias, T. Ning, O. Bock, J. Guijarro, F. Alshawaf, M. Hoseini, A. Quarello, E. Lebarbier, B. Chimani, V. Tornatore, S. Zengin Kazancı, J. Bogusz (2020). Homogenizing GPS integrated water vapor time series: benchmarking break detection methods on synthetic datasets. Earth and Space Science, 7, e2020EA001121. https://doi.org/10.1029/2020EA001121

Venema, V., Mestre, O., Aguilar, E., Auer, I., Guijarro, J. A., Domonkos, P., et al. (2012). Benchmarking monthly homogenization algorithms. Climate of the Past, 8, 89–115. https://doi.org/10.5194/cp‐8‐89‐2012