the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Nov 2025
Comparison of water vapour products retrieved from Metop-A by Radio Occultation, Infrared and Microwave systems
Abstract. Water vapour (WV) is an essential climate variable (ECV) and different Earth Observing Systems (EOS) have been used to monitor and characterise its distribution, transportation and interplay in different phenomena. Metop-A satellite carried since 2006 four of such systems: (i) Infrared Atmospheric Sounding Interferometer (IASI), (ii) Advanced Microwave Sounding Unit (AMSU-A), (iii) Microwave Humidity Sounder (MHS), and (iv) Global navigation satellite system Receiver for Atmospheric Sounding (GRAS) performing radio occultation (RO) measurements. These systems operate at different frequencies, with different acquisition geometries, and also use different retrieval schemes to obtain vertically resolved WV profiles in the troposphere. Therefore, they provide independent measurements of WV, each with its limitations and strengths. Their characterisation via comparison with other datasets is important to estimate their systematic differences and uncertainties, which must be known when it comes to their use as climate data records (CDR) in climate monitoring and change studies. In this study, the water vapour product part of the Rutherford Appleton Laboratory (RAL) Infrared and Microwave System (IMS) scheme used in the European Space Agency (ESA) Water Vapour Climate Change Initiative (WV_cci) is compared to GRAS-RO WV data, part of Radio Occultation Meteorology Satellite Application Facility (ROM SAF) CDR, v1.0. The comparison uses ERA-Interim analysis and the Global Climate Observing System (GCOS) Reference Upper-Air Network (GRUAN) as references and presents the systematic differences in different latitude bands, gauges the influence of cloud contamination on the statistics, as well as the differences between profiles over land and water, and during day and nighttime. Results in the lower troposphere (LT) show a significant difference between the data sets, where RAL IMS is wetter than ERA-I and GRUAN, especially in mid and high latitudes and over water (up to 14 % ppmv) regardless of clouds, whereas GRAS-RO is drier than the references (5.3 % ppmv at maximum w.r.t. GRUAN and during night). In the upper troposphere (UT), both data sets are drier than ERA-I (16.8 % ppmv, RAL IMS; 7.2 % ppmv, RO) and wetter than GRUAN (up to about 20 % ppmv). In terms of variability, smoothing GRAS-RO, ERA-I and GRUAN with RAL IMS WV averaging kernels (AK) reduces the magnitude of their median absolute deviation (MAD) and increases their similarities (LT: RO, 8.78 % ppmv; RAL IMS, 12.75 % ppmv. UT: up to 36.33 % ppmv).
- Preprint
(945 KB) - Metadata XML
-
Supplement
(140 KB) - BibTeX
- EndNote
Status: open (until 19 Feb 2026)
-
RC1: 'Comment on egusphere-2025-5578', Anonymous Referee #1, 22 Dec 2025
reply
-
AC1: 'Reply on RC1', Vinícius Ludwig Barbosa, 29 Jan 2026
reply
The authors appreciate the reviewer's comments. Below are our replies, which will be incorporated accordingly into the revised version of the manuscript after the open discussion closes.
In the first paragraph, which summarises our study:
"The authors find that RAL IMS is wetter than the reference datasets in the lower troposphere but drier in the upper troposphere..."
We would like to emphasise that RAL IMS is wetter than GRAS-RO and the references (ERA-Interim and GRUAN) in the lower troposphere. In the upper troposphere (UT), RAL IMS is drier than GRAS-RO and ERA-Interim, but wetter than GRUAN.
"...while GRAS-RO is drier than the references along the entire retrieved water vapor profile."
GRAS-RO is drier than ERA-Interim along the entire retrieved WV profile. However, it is wetter than GRUAN at the top of the mid troposphere (MT) (~500 to 400 hPa) and in the upper troposphere, to a similar level as seen in RAL IMS WV data.
1. Major comments
1) The goal of our study is to provide a comparison between RO and RAL IMS WV data globally, in contrast to GRUAN data, which has limited coverage. To do so, we have followed the methodology described in Trent et al. (2023) to a large extent. Nevertheless, differences between the methodologies are summarised below:
- Data sets: The matchup datasets in Trent et al. and our study were generated independently. We generated a dataset whose matchups result from a search for collocations between all nominal GRAS occultations and scenes in the RAL IMS dataset, v2.1. Therefore, the results shown in Figure 3-5 are not limited to GRUAN stations, as in Trent et al. We will add this point to Section 4.1.
Results for the comparison against GRUAN are based on a subset of RO-IMS matchups, whose RO profiles are part of the dataset of RO-GRUAN matchups prepared internally at DMI prior to this study. Every profile in a triplet satisfied the spatial and temporal collocation criteria among its pairs. A consequence of enforcing triplets was a smaller dataset than that assessed in Trent et al.. These details will be added in Section 4.2.
Also, our analysis did not include Lamont and La Réunion stations. It did include data from the station in Singapore. This is already stressed in Appendix C.
- Collocation criteria: We have extended the spatial criteria from 100 km to 300 km. This change accounted for the distance covered by an occultation event and maximised the number of matchups. See lines 190-194.
Particularly for the first part of the analysis, matchups satisfied an additional requirement: the maximum difference in surface altitudes between pairs must be no more than 10 hPa (~100 m). See lines 205-207.
- Data preparation and statistical analysis: In Trent et al. (2023), GRUAN profiles were downsampled to the RAL IMS 101-level grid, then filtered using their RAL IMS averaging kernel pairs. Results were then presented at weighted layers, adding a third stage of smoothing to the GRUAN data. We followed the first two stages in our methodology. However, we skipped the last one and presented the statistics in the 101-level grid instead. For GRAS-RO data (part of ROM SAF CDR v1), WV profiles were upsampled from 60 levels to the RAL IMS 101-level grid. This discussion will be added in Section 3.2. The consequence of presenting the results as weighted layers was already addressed in lines 344-358.
Further, we added a constraint that the collocation pair or triplet must share the same feature in the analyses assessing data subsets (land & ocean, day & night), e.g., all measurements during the daytime. See lines 309-311 and 353-354.
- Findings: Regarding the first part (Section 4.1), our results provide a more extensive analysis of the material presented in Trent et al. (Figure 1, rightmost panels). Our results differed partly due to the amount of data (one day versus 9.5 years) and to the reference (ERA-Interim versus ERA-5). However, they agreed to show a wet bias in RAL IMS from the surface up to about 750 hPa. We will stress this agreement in Section 4.1.
Regarding the second part, our results are comparable to Section 4 in Trent et al., except for the breakdown of statistics at specific GRUAN sites and the assessment of temperature data. In the accumulated statistics (“Global”), we confirmed the trends and magnitudes of the biases and variability reported in Trent et al., and we showed that wet bias in the lower troposphere (LT) likely comes from profiles in high latitudes (northern hemisphere). Also, in LT, we showed that RO WV profiles are consistent across latitudes and show a minor dry bias relative to GRUAN. See lines 323-330.
Our results differ in the share of day and night biases. We showed that LT biases are roughly equal between day and night, whereas Trent et al. observed a bias only during the daytime. This statement is available in the manuscript in lines 355-360, where a correction is needed:
“...even though the wet bias is also observed in daytime (nighttime) profiles – a pattern not observed in Trent et al. (2023).”
In the mid and upper troposphere, the largest share of wet bias in RAL IMS occurs in nighttime profiles, according to our results. In Trent et al., this bias is nearly equally distributed between day and night. Daytime matchups accounted for about 83% of our dataset. In Trent et al., daytime matchups were 44% of their dataset. Conclusions regarding MT and UT statistics will be added in lines 361-366.
In assessing cloud fraction dependence on biases, we used only cloud fraction in our data selection and did not use the BTD flag, as in Trent et al. The cloud fraction dependence was evaluated only in the first part, since the dataset is larger and provides a clearer picture of global dependencies. Our results agreed that the LT wet bias is not dependent on cloud fraction (see lines 301-302). In UT, both show a relation between wet bias and high cloud fractions, and/or high cloud-top heights (>3 km) (see lines 302-303).
2) The theory of AK filtering is described in detail in Trent et al. (2023) and in Rodgers et al. (2003). For this reason, we did not allocate more space to describe it in our manuscript. We will include one more reference:
- Maddy, E. S. and Barnet, C. D.: Vertical resolution estimates in version 5 of AIRS operational retrievals, IEEE T. Geosci. Remote, 46, 2375–2384, 2008.
Figures 1 and 2 do not reproduce the results of Trent et al. They characterise the RO WV data in terms of cumulative degrees of freedom (related to vertical resolution) and the vertical range over which the RO data is most valuable (Figure 1). Figure 2 illustrates the general outcome of smearing RO WV profiles using RAL IMS AKs.
While the application of AKs to high-resolution profiles is important for establishing a fair comparison with IMS profiles, we observed that the RAL IMS AKs introduced biases in profiles with higher resolution than IMS. Figure 3 shows that applying AKs introduces substantial biases in the comparison between RO and ERA-I. Figure 6 shows that applying AKs introduces substantial biases of opposite sign in the comparison between RO and GRUAN. This observation leads us to be a bit careful with interpreting high-resolution profiles modified with AKs. We note that the RO profiles have a smaller vertical footprint than the ERA-I, and that the GRUAN profiles have a vertical footprint even smaller than that of the RO profiles. We speculate that biases introduced by the AKs are linked to the vertical footprints of the smoothed datasets, but we do not have an analytical result to support this. The introduction of these biases is why we opted not to draw too many conclusions from the results based on AK-smoothed high-resolution data.
3) It is a good suggestion to choose the presumably most accurate and high-resolution dataset as the reference. In this case, we consider the well-calibrated GRUAN radiosondes as the best candidate. GRUAN has the downside of being a very sparse dataset. Hence, we will rewrite the conclusions such that the rationale is that GRUAN is the reference and RO has a moderate constant bias structure, i.e., slightly dry in the lower and mid troposphere and moist in the upper troposphere. The IMS biases are then evaluated globally relative to the RO data, which has known biases. In the tropics, we are more reluctant to draw conclusions about RAL IMS data, since only a relatively small number of matchups were available due to a limited number of GRUAN stations. Nevertheless, the evaluation of RAL IMS against RO is the main objective of the paper, and there is a large number of collocations at low, mid and high latitudes (Section 4.1).
4) We will add the discussion presented in point 2 to the conclusions. In the current version of the manuscript, we can state that the performance goal of 10% ppmv globally holds relative to GRUAN. However, this goal does not hold at high latitudes (northern hemisphere) or at night.
2. Minor comments
Both comments will be addressed in the revised version of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-5578-AC1
-
AC1: 'Reply on RC1', Vinícius Ludwig Barbosa, 29 Jan 2026
reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 243 | 75 | 25 | 343 | 33 | 22 | 25 |
- HTML: 243
- PDF: 75
- XML: 25
- Total: 343
- Supplement: 33
- BibTeX: 22
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper analyzes water vapor data from the Rutherford Appleton Laboratory (RAL) Infrared and Microwave System (IMS) against GRA- RO data from Metop-A. ERA-Interim and the Global Climate Observing System Reference Upper-Air Network (GRUAN) are used as references. The authors find that RAL IMS is wetter than the reference datasets in the lower troposphere but drier in the upper troposphere, while GRAS-RO is drier than the references along the entire retrieved water vapor profile. The authors also smooth GRAS-RO and both reference datasets using RAL IMS water vapor averaging kernels, and find that this reduces the magnitude of their median absolute deviation and increases their similarities.
Major comments:
Minor comments: