the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Apr 2026
A multi-year dataset of integrated water vapor derived from shipborne GNSS observations collected aboard eight French research vessels during oceanographic campaigns (2015–2024)
Abstract. In the context of climate change and the growing need for improved observations of the atmospheric water cycle, measurements of atmospheric water vapor over the global ocean remain scarce compared with those available over land. This observational gap can be addressed using Integrated Water Vapor (IWV) derived from shipborne Global Navigation Satellite System (GNSS) observations, which provide a robust and well-established method for monitoring atmospheric moisture over the oceans.
This study presents a shipborne IWV dataset obtained from the processing of raw data collected by GNSS antennas installed on research vessels. The dataset benefits from substantial support and data access provided by Genavir, the operator of the French Oceanographic Fleet (FOF), and the Ifremer archive department SISMER. It is based on oceanographic campaigns conducted worldwide by eight vessels over a ten-year period (2015–2024), representing a total of 6,427 campaign days in both offshore and coastal regions.
After describing the methodology used to derive IWV from raw GNSS observations and to remove spurious measurements through a screening procedure, the dataset is evaluated through comparisons with the ERA5 reanalysis and satellite radiometer measurements from Remote Sensing Systems. These comparisons yield mean differences of (0.3 ± 2.0) kg m−2 and (−0.4 ± 1.8) kg m−2, respectively.
To further quantify the inherent uncertainty of the shipborne IWV retrieval, the dataset is cross-validated using instances where two vessels were within 50 km of each other. This comparison results in an estimated uncertainty of 0.96 kg m−2, demonstrating the suitability of the dataset for climate studies.
Local discrepancies identified in these comparisons are discussed, highlighting limitations in each dataset considered.
The GNSS-derived IWV dataset is intended to be updated annually to support long-term monitoring of atmospheric water vapor over the global ocean.
The IWV estimates are available at https://doi.org/10.25326/876 (Panetier and Bosser, 2026) through the AERIS data center (https://en.aeris-data.fr/, last access: 20 March 2026), and currently span the period from 2015 to the end of 2024.
- Preprint
(9367 KB) - Metadata XML
-
Supplement
(22 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2026-223', Jungang Wang, 12 May 2026
-
AC1: 'Reply on RC1', Aurélie Panetier, 06 Oct 2026
We would like to express our sincere gratitude to the reviewer for the time and attention devoted to our manuscript. We were genuinely moved by the enthusiasm and interest shown for this work, and we are grateful for such a thorough and encouraging reading of the paper and for these constructive comments. We address each of these comments in detail below.
1. We appreciate that you have pointed out this missing but valuable information. This information has been added to the data availability section : "The GNSS antenna binary data files are available upon request through the IFREMER Data Center Portal SISMER https://data.ifremer.fr/Centre-de-donnees/Contact."
2. Thank you for this suggestion. The uncertainties are provided in the dataset as sig_ztd, sig_zhd, sig_Tm and sig_iwv for the ZTD, ZHD, Tm and IWV standard deviations respectively. The content of the netcdf files provided have been added to the data section of the paper: "For each vessel, the dataset provides the following variables: longitude and latitude (°), ellipsoid height and mean sea level height (m), zenith troposphere delay, zenith hydrostatic delay, and zenith wet delay (m), mean temperature (K), integrated water vapor content (kg/m²), and the formal errors associated with the zenith troposphere delay, zenith hydrostatic delay, zenith wet delay, mean temperature, and integrated water vapor content."
3. Thank you for this suggestion. The dataset is meant to be a valuable dataset providing information on its own, and that will be fed with futur campaigns to come from the FOF. Then, the inclusion of external data is not suitable directly in the AERIS dataset, especially because this study will not automatically be conducted in the future, especially not with the radiometer comparison. However, we provide ERA5 and radiometer time series aas supplementary material, in order to help the reviewers address this interesting point, at this location: https://doi.org//10.5281/zenodo.23167143.
Citation: https://doi.org/10.5194/essd-2026-223-AC1
-
AC1: 'Reply on RC1', Aurélie Panetier, 06 Oct 2026
-
RC2: 'Comment on essd-2026-223', Anonymous Referee #2, 21 Jul 2026
The manuscript "A multi-year dataset ..." submitted by A. Panetier presents a very valuable compilation of ship-based water vapor products, filling an essential gap in the usually land-based GNSS product series. The manuscript is well-written and the results are supported by reasonable validation. In terms of the layout the authors could consider to merge some of the shorter paragraphs which partly cover only two sentences (see for example section 2.2.2). Please find some detailed comments below.
- l118 "antenna height", how is this defined, above sea level or related to some marker?
- l122, was the number of satellites per epoch considered?
- l147 I guess "observations" refer now to ZTDs?
- l190 An additional study for this aspect is potentially https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2024GL113140
- l224 I would not consider ERA5 as "technique"
- l238 Are these wet and dry biases related to remaining systematics and data quality?
- l270 Would it be possible to show a distance vs difference plot?
- Figure 9, might be interesting to add ERA5-based IWV here.
- l289 I suggest to check also the GNSS data quality indicators and the processing statistics.
- Figure 11, might be interesting to add the IWV for GNSS station NRMD here.
- Figure 15, I wonder whether this constant offset is related to any issue in the coordinate estimation or some height offset?
Citation: https://doi.org/10.5194/essd-2026-223-RC2 -
AC2: 'Reply on RC2', Aurélie Panetier, 06 Oct 2026
We thank sincerely the reviewer for their careful reading of the manuscript. The reviewer's comments are both pertinent and constructive, and engaging with them has allowed us to improve the manuscript and clarify several aspects of the dataset that will benefit its future users. Our point-by-point responses are given below. Changes made to the manuscript are indicated accordingly.
General comment: Thank you for providing insights to improve the layout. We have read the whole article while taking care to use one paragraph to develop one idea instead of cutting it. We hope this suits the reader better.
l118: We have clarified this point. The term now reads "antenna height formal error", since we are referring to the formal error of the antenna height estimate, used here as a quality indicator of the processing, rather than to the antenna height itself. The antenna height here, is related to the ellipsoid.
l122: This was not considered in the present study. However, we agree that it would be valuable information, and we intend to use it to refine the screening criteria in future work. We are now addressing this point in the Discussion section of the article.
l147: Yes. We have revised the text accordingly to be more precise on this point.
l190: We thank the reviewer for pointing us to this relevant reference. Yuan et al. (2025, Geophysical Research Letters) identify artificial discontinuities in ERA5-derived zenith total delays (ZTD) at 09:00 and 21:00 UTC, based on a comparison with a decade of GNSS data from more than 6,000 stations worldwide. Our own dataset is far too sparse, both in number of stations and in time span, to carry out a comparable assessment of this diurnal error in ERA5. Nevertheless, the existence of such an error is a valuable additional argument: even though we cannot verify it with our data, it illustrates that ERA5 is not free of systematic errors, and that this could also contribute to the differences we observe in our comparisons. We have therefore added this reference to the manuscript.
l224: We agree and have changed "technique" to "dataset" throughout the relevant passage.
l238: Thank you for this comment. We investigated several possible explanations for the differences in wet and dry biases between vessels. Looking more closely at the data, we found that the larger wet bias and standard deviation observed for Alis are related to a data quality issue linked to the number of visible satellites: on the days concerned, and even on adjacent days where the effect is less obvious from the other indicators alone, the number of observations available is markedly lower. This connects to the reviewer's earlier comment (l. 122) on the number of satellites per epoch, which we now discuss explicitly in this context. We show that this occurs relatively rarely across the dataset, which is reassuring, but that it affects Alis more than the other vessels. In particular, for Atalante, filtering on the ZTD formal error (sigma) is effective at removing epochs with a low number of observations, whereas this same filtering is less effective for Alis, so that degraded, low-observation epochs are not as thoroughly screened out. We also examined a possible dependence on latitude: the mean bias and standard deviation are larger between 0 deg and 30 deg latitude (0.52 +/- 2.36 kg.m-2, N = 67,497) than between 30 deg and 60 deg latitude (0.12 +/- 1.63 kg.m-2, N = 75,259), which may further contribute to the larger discrepancies observed for Alis, given that it operates in the tropical Pacific. In contrast, seasonality and vessel type do not appear to explain the differences, since all vessels operate across the different seasons covered by the dataset, and Alis and Antea are of similar (semi-offshore) vessel type. These considerations have been added to the revised manuscript, with further discussion in Section 4.1.3.
l270: Thank you for the suggestion. We initially considered a distance-versus-difference plot; however, combining the data from both vessels in a single figure proved difficult to interpret. Moreover, the IWV differences do not show a clear correlation with distance such as seen in figure attached as supplement to this answer, so such a plot did not add clear additional insight. We therefore chose instead to represent Delta-IWV alongside the crossing time series, with an additional subplot showing the distance between the two vessels underneath (Fig. 9). This allows a direct, more interpretable comparison of the temporal evolution of the IWV differences with the inter-vessel distance.
Figure 9: Thank you for the suggestion. We agree that adding ERA5-based IWV provides a useful point of comparison with the vessel-derived IWV. We have produced a revised version of the figure showing ERA5-based IWV at both vessel locations alongside the corresponding vessel-derived IWV. To keep the figure legible despite the four resulting curves, GNSS-derived IWV is shown as the vessel-specific colored curves, while ERA5-based IWV interpolated at the GNSS measurement locations is shown as black-outlined circles filled with the corresponding vessel color. This preserves the requested comparison while keeping the two vessels and the two data sources clearly distinguishable.
l289: Thank you for this suggestion. We checked the GNSS data quality indicators for the three days in question, including the MP1 and MP2 multipath statistics and the number of cycle slips identified from the GF, MW, and IOD(L1) combinations. No significant degradation of these indicators was observed on the middle (anomalous) day: the MP1 and MP2 standard deviations remain very similar across doys 178, 179, and 180, and the cycle-slip statistics show no clear increase on the anomalous day. These additional diagnostics therefore do not point to any clear degradation of GNSS data quality or processing performance that could explain the observed anomaly. We also examined the formal error of the ZWD estimation, shown in the lower panel of Fig. 10, which reflects the quality of the computation. It is slightly higher on the nomalous day, but it is already high on the other days. We then addressed the question of the number of observables, and found that it was weak on this whole period for Alis, which may explain as well its lower quality compared to other vessels. This result led to a new figure in the revised manuscript: Fig. 11. All this have been added to the article and provides valuable insights into the data quality.
Figure 11: Thank you for the idea. We have added the IWV time series for station NRMD, that we computed using GipsyX, to the figure. It is now Fig. 12. We also added a map showing the location of the different antennas involved in the comparison.
Figure 15: Thank you for this comment that pushed us to look deeper into this issue. This figure is now Fig. 16. All comparisons are made between IWV datasets that are either directly retrieved at, or extrapolated to, sea surface level, so all values are nominally at the same height. However, this is the only crossing in the dataset involving Marion Dufresne, and it would have been possible that the offset stems from an issue in the Marion Dufresne antenna height estimation, since no comparable offset is observed in other crossings of Pourquoi Pas? with other vessels. The Marion Dufresne antenna is indeed known to have issues (see Bosser et al., 2022), but not on this period. This is why we investigated this case further by comparing it against ERA5, and we have now also added the nearby station LEPO to the comparison. The picture that emerges is in fact more complex than initially stated. We have added the LEPO time series to the figure and revised the corresponding text accordingly. Our conclusion to the observations with LEPO is that the area was highly heterogenous on this day, the difference between the two vessels 2km apart is then due to different atmospheric states above their antennas. To specifically answer your point, we have compared the ZHD values of Pourquoi Pas? and Marion Dufresne, and they appear to be strictly identical, thus there is no shift in Marion Dufresne antenna height above sea level. We added a map showing the location of the different antennas involved in the comparison to this figure.
Data sets
Integrated Water Vapor contents from GNSS observations collected onboard research vessels of the French Oceanographic Fleet Aurélie Panetier and Pierre Bosser https://doi.org/10.25326/876
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 466 | 180 | 129 | 775 | 159 | 151 | 116 |
- HTML: 466
- PDF: 180
- XML: 129
- Total: 775
- Supplement: 159
- BibTeX: 151
- EndNote: 116
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript presents a valuable multi-year dataset of integrated water vapor (IWV) derived from shipborne GNSS observations collected aboard eight French research vessels during oceanographic campaigns between 2015 and 2024. The topic is timely and relevant, as observations of atmospheric water vapor over the oceans remain sparse compared with land areas, while such measurements are essential for climate studies, numerical weather prediction, satellite validation, and the investigation of ocean–atmosphere interactions.
The main strength of the manuscript is the systematic construction of a long-term and geographically extensive IWV dataset from opportunistic shipborne GNSS observations. The authors provide a clear description of the GNSS processing strategy, the screening procedure, and the conversion from ZTD/ZWD to IWV. The dataset is carefully evaluated using independent reference data, including ERA5 reanalysis, satellite microwave radiometer products, and inter-vessel comparisons when two ships were located within 50 km of each other. The reported agreement with ERA5 and RSS radiometer products, as well as the internal consistency shown by vessel-crossing comparisons, demonstrates the overall quality and usefulness of the dataset.
I particularly appreciate that the authors do not limit the validation to global statistics, but also investigate specific outlier cases in detail. The discussion of discrepancies related to island/coastal regions, ERA5 interpolation limitations, latitude-dependent differences, cloud-cover-dependent radiometer biases, and occasional vessel-specific GNSS issues provides useful insight into both the strengths and limitations of the dataset. This makes the manuscript more informative and increases confidence in the data product.
Overall, I find the dataset scientifically useful and well suited for publication in Earth System Science Data. It provides an important contribution to oceanic water vapor observations and has clear potential for applications in satellite product validation, reanalysis assessment, climate monitoring, and future data assimilation studies. I therefore support publication of the manuscript after minor revisions. Some points could be further clarified to improve the manuscript and enhance the usability of the dataset.
Specific comments
1. It would be useful if the authors could clarify the availability of the original GNSS observation data, preferably in RINEX/RNXO format, if these data are publicly accessible. Although the derived IWV product is the main dataset of this paper, access to the original GNSS observations would further improve reproducibility and would be valuable for users interested in reprocessing the data with alternative strategies.
2. I suggest that uncertainty information for the GNSS-derived PWV/IWV product, or at least for the intermediate ZWD/ZTD estimates, be included in the dataset. If this is not feasible for the current release, the authors could consider adding such information in future updates. This would be helpful for data assimilation, validation studies, and quantitative comparison with other water vapor products.
3. The authors may also check whether the comparison datasets or collocated comparison results used in the manuscript, for example the interpolated ERA5 IWV values and the matched RSS radiometer values, can be made available together with the GNSS-IWV product. This is not essential, since ERA5 and RSS are already available from their respective providers, but providing the collocated comparison data would facilitate reproducibility of the validation results and make the dataset more convenient for users.