Comparison of M10 and M20 Meteomodem radiosondes relative humidity measurements with ECMWF ERA5 above France: focus on the upper troposphere

Diarra, Sidy; Montoux, Nadège; Baray, Jean-Luc; Keckhut, Philippe; Dupont, Jean-Charles; Farah, Antoine; Alraddawi, Dunya

doi:10.5194/egusphere-2026-482

Preprints

https://doi.org/10.5194/egusphere-2026-482

Preprints

03 Mar 2026

| 03 Mar 2026

Status: this preprint is open for discussion and under review for Atmospheric Measurement Techniques (AMT).

Comparison of M10 and M20 Meteomodem radiosondes relative humidity measurements with ECMWF ERA5 above France: focus on the upper troposphere

Sidy Diarra, Nadège Montoux, Jean-Luc Baray, Philippe Keckhut, Jean-Charles Dupont, Antoine Farah, and Dunya Alraddawi

Abstract. Accurate knowledge of the relative humidity (RH) in the troposphere is important for predicting cloud formation, particularly in the upper troposphere where contrails can form and contribute to global warming. However, it is difficult to predict their formation due to the lack of precise RH measurements at these altitudes. This paper compares RH data from Meteomodem radiosondes (M10 and M20) acquired over a 5-years period (2020-2024) at the Trappes and Nîmes meteorological stations in France with ECMWF ERA5 analyses, with a focus on the upper troposphere. For Trappes, two datasets exist: one processed operationally by Météo France (MF) and a second processed using the GRUAN standard. Whatever the processing is, Meteomodem radiosondes RH values are on average higher than ERA5 ones, by about 2 % at 800 hPa up to 10 % at 200 hPa. The operational MF processing generally gives higher RH than the GRUAN processing. The median difference between both processing methods is lower than 2.2 % for pressures higher than 300 hPa and is maximum for lower pressures and nighttime measurements, the GRUAN processing showing more consistency between daytime and nighttime measurements. The evolution of MF processing over time does not seem to affect the comparison. The major differences observed between the relative humidities measured by the sondes and those provided by the ERA5 reanalysis are between 200 and 300 hPa. First, ERA5 indicates more occurrences of RH below 40 % than the sondes. Second, the sondes indicate supersaturation conditions (~20 %) more frequently than ERA5 (11 %), probably due to the cloud parameterization in the IFS model, which fixes the RH at 100 % as soon as a cloud forms, in agreement to the higher occurrence of saturation conditions observed by ERA5 in this study. A first comparison of the results obtained at Trappes and Nîmes between the year 2020 and the years 2022, 2023 and 2024 shows no major differences, suggesting that the switch from M10 to M20 sondes in March 2021 at Nîmes does not significantly affect the ability to combine RH data for long term trends. However, more detailed investigations are required to assess finer differences. Finally, this study underlines the need to continue efforts to assess the quality of RH measurements in the upper troposphere and to improve cloud parameterizations in the model to increase supersaturation frequency in the upper troposphere as observed by the sondes.

Received: 27 Jan 2026 – Discussion started: 03 Mar 2026

Competing interests: Author Antoine Farah is employed by the Meteomodem company.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Sidy Diarra, Nadège Montoux, Jean-Luc Baray, Philippe Keckhut, Jean-Charles Dupont, Antoine Farah, and Dunya Alraddawi

Status: open (until 14 May 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-482', Anonymous Referee #1, 25 Mar 2026 reply

This paper compares radiosonde relative humidity measurements in the extratropical upper troposphere using Meteomodem M10 or M20 with collocated relative humidities from ERA5 over an extended period. Results show systematically higher relative humidities in the radiosondes compared to ERA5, particularly around 200 hPa. Two methods of raw data processing are compared, where GRUAN processing leads to systematically lower relative humidities than the Meteomodem processing at some altitudes. Little difference is found between M10 and M20 radiosondes.
Major issues: The paper does pay too little attention to the international radiosonde intercomparison in Lindenberg (Dirksen et al. 2024). This paper is cited and it is mentioned that M20 has a wet bias during nighttime compared to the standard defined in Dirksen's report. It seems highly likely that much of the presented discrepancy between ERA5 and M10/20 radiosondes is related to that wet bias. The wet bias can be partly reduced with GRUAN processing, which is the most important result of this study. The authors however attribute the discrepancy to potential errors in ERA5 (too little supersaturation over ice events due to deficient parameterization). To me this is highly doubtful, first because ERA5 relative humidity is reported relative to water, and it is unclear to me how the authors converted this relative humidity to relative humidity with respect to ice. There are uncertainties involved due to interpolation and averaging in ERA5 boxes at least.
Parts of the conclusions are written rather vaguely, giving the impression of inconclusive results. This does not help the reader at all. It may help to give uncertainty estimates instead of formulations such as "shouldn't prevent climatological studies". At least some of these uncertainty values can be found in the main body of the paper. It is important to state these also in the conclusions since often only the conclusions are read.
MInor revisons:
ERA5 (Hersbach et al. 2020) should be cited a lot earlier at the first occurence in the introduction.
lines 262 and 270/271: This appears redundant, consider shortening this paragraph
387: Data processing evolution is a bit awkward in this context. It is more stepwise changes. Thus I would say "changes" or "upgrades"
line 213f: It is not entirely clear to me if this (taking RH from ERA5, and T from the sondes) could be a reason for the different supersaturation frequencies in ERA5 and the radiosondes. This should be elaborated with some sensitivity experiments/uncertainty estimates
line 464: daytime! radiosoundings
line 506: larger (not greater)
lines 524, 548: IFS
line 540: Is there a public document describing this campaign
lines 529ff: as noted above: the conclusions are formulated very vaguely here ("shouldn't prevent", "appears reproducible"). It would be better to have uncertainty ranges

Reply

Citation: https://doi.org/10.5194/egusphere-2026-482-RC1
- AC1:
  'Reply on RC1', Nadège Montoux, 16 Apr 2026 reply
  We would like to thank Referee #1 for his/her review. At this stage of the review process, we provide here a response to the main comment raised by the referee. A revised version of the manuscript and a detailed point-by-point response addressing all comments and requests, will be submitted once the interactive public discussion phase has been completed.
  Referee comment: The paper does pay too little attention to the international radiosonde intercomparison in Lindenberg (Dirksen et al. 2024). This paper is cited and it is mentioned that M20 has a wet bias during nighttime compared to the standard defined in Dirksen's report. It seems highly likely that much of the presented discrepancy between ERA5 and M10/20 radiosondes is related to that wet bias. The wet bias can be partly reduced with GRUAN processing, which is the most important result of this study. The authors however attribute the discrepancy to potential errors in ERA5 (too little supersaturation over ice events due to deficient parameterization). To me this is highly doubtful, first because ERA5 relative humidity is reported relative to water, and it is unclear to me how the authors converted this relative humidity to relative humidity with respect to ice. There are uncertainties involved due to interpolation and averaging in ERA5 boxes at least.
  Authors answer: In the present paper, as mentioned in section 2.1, ERA5 relative humidity is not systematically defined with respect to liquid water; instead it is defined with respect to liquid water or ice depending on temperature T. If T is lower than 250.16 K, RH is relative to ice and if T is higher than 273.16 K, RH is relative to liquid water. Between 250.16K and 273.16K, a combination of both is used as described in equations 2 to 5. At the altitudes where the largest differences between radiosondes and ERA5 are observed (200-300 hPa), the temperature is always below 250.16 K and RH is defined with respect to ice.
  In this paper, significant effort has been devoted to homogenizing RH calculations (see section 2), allowing a consistent comparison between radiosondes and ERA5 RH without biases induced from different formulations of saturation water pressure.
  Despite this homogenization, differences in RH between M10/M20 sondes and ERA5 remain. We agree that these difference may partly be attributed to wet bias previously reported by Dupont et al. (2020) for M10 and Dirksen et al. (2024) for M20 sondes. However, our results indicate that this cannot be the only explanation for the following reasons :
  the wet bias observed for M10 in this study is larger than previously reported by Dupont et al. (2020).
  
  Dupont et al. (2020) showed that differences relative to Vaisala RS92 RH values are within 2% (nighttime) and 5% (daytime) for the GRUAN processing, and within 6% and 9% for operational processing. In contrast, the bias shown in figure 1 of the present study increases with altitude and reach 10% at 300 hPa (~9-10 km) for both daytime and nighttime measurements and for both processing methods.
  the bias cannot be explained by systematic errors as suggested by Dirksen et al., (2024).
  
  By comparing the mean measurement error and standard deviation of the individual measurement errors (Figures L87 and L88 in Dirksen et al., 2024), these authors showed that the wet bias is mainly due to systematic errors, at least between 4 and 12 km for both daytime and nighttime measurements. However, in our analysis of the 200-300 hPa layer, the difference between radiosondes and ERA5 RH increases for ERA5 RH values above 90% with a bias between 20 and 25%. M10 sondes show more frequent supersaturation than ERA5, which shows a high occurrence of RH values close to 100 %. At these altitudes, the difference may result not only from the known wet bias in the M10 data, but also from a dry bias in ERA5, as reported in comparisons with lidar data (Alraddawi et al., 2025) and MOZAIC aircraft data under supersaturation conditions (Gierens et al., 2020). In addition, laboratory experiments and in flight comparisons conducted under cloudy conditions generally show a good agreement between M20 sondes and references instruments (Figures 10.24 and 10.25 of Dirksen et al., 2024). Only a few measurements indicate an underestimation of RH by M20 radiosondes, which would suggest a reduced wet bias in cloudy conditions, contrary to what is observed here.
  Regarding uncertainties related to interpolation and averaging within ERA5 grid boxes, we minimized interpolation uncertainties by performing 3D interpolation using the radiosonde trajectory and by using the highest ERA5 resolution available (0.125° horizontal resolution and on 137 verticals model levels). Nevertheless, radiosonde measurements are point measurements while ERA5 provide grid boxes averages. Within each grid box, RH can be expressed as a cloud fraction weighted mean : RH = C × 100 + (1 − C) × RH_clear.
  with C is the cloud fraction and RH_clearis RH in clear sky condition. However, given the large number of profiles used in this study, there is no reason to expect a systematic sampling bias toward one or the other. Therefore, this effect is unlikely to explain the observed differences.
  Instead, the difference is more probably related to the microphysical parametrization than spatial resolution. As mentioned by Gierens et al., (2020), “When a grid box contains a cloud, the humidity in the cloudy part of the box is immediately (within one time step) reduced to saturation” and “When the humidity in the clear part of the grid box, RH_i,clear, increases and surpasses a critical value, the cloud fraction C increases as well; this balancing effect inhibits the increase of grid mean humidity to observed maxima.” These mechanisms can lead to an underestimation of RH in cloudy conditions, consistent with our results (figures 2 and 3).
  Additional sentences will be added in the revised manuscript to better discuss the observed bias and reference to the Dirksen et al., (2024) publication will also be added in this part.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2026-482-AC1
RC2: 'Comment on egusphere-2026-482', Anonymous Referee #2, 22 Apr 2026 reply

The manuscript "Comparison of M10 and M20 Meteomodem radiosondes relative humidity measurements with ECMWF ERA5 above France: focus on the upper troposphere" presents the results of a comparison of the relative humidity values in the upper troposphere (200-300hPa), a part of the atmosphere relevant for contrail formation, from ERA5 reanalysis data versus operational in situ observations with the Meteomodem M10 and M20 radiosondes.

In addition the authors investigate the differences between manufacturer-processed data and data processing developed for GRUAN.

The authors find a clear dry bias for ERA5 data at high RH values, which is attributed to the fast cloud formation scheme in the numerical weather model that basically suppresses supersaturation.

For lower RH values, far away from saturation, the difference between ERA5 and radiosonde RH data is greatly reduced.

Furthermore, the authors report a relatively small dry bias for the manufacturer-processing, and also address the potential effects of change of the sounding model (M10 to M20) to the data.
A major deficiency of the manuscript is that there appears to be no clear common thread or storyline. It feels like the collage of loosely-connected topics and results that are lumped into one document.

As a consequence, the various topics are not covered with the necessary depth and detail, which stands in the way of reaching appropriate scientific value of the work presented in the manuscript.

High-level comments
The main motivation for this study provided in the introduction is the need for accurate humidity observations in the upper troposphere connected to the alleged overestimation of cloud formation by ERA5, as is indicated by the description of ERA5's cloud formation scheme by Tompkins et al. 2007.

To get an idea of the importance of this issue it would be good if the authors can provide additional background information in the form of references to relevant studies. Are there observations that support this alleged overestimation of the cloud cover by ERA5, do for example satellite-based cloud observations show discrepancies with ERA5?

Suitable references in this regard may be Wolf et al. (https://doi.org/10.5194/acp-25-157-2025) or Hildebrandt et al. (https://doi.org/10.5194/egusphere-2025-3048)
The description of ERA5 is a bit coarse. The authors could provide some information on the numerical model that ERA5 employs (the IFS). For example does the model include atmosphere-ocean coupling, what are the constraints, which schemes are used, what are performance-wise the strengths and weaknesses of the model?
At some locations in the manuscript the description of the Meteodem radiosonde systems is reads too much like a marketing brochure, please keep it neutral and factual.
The common parameter for the comparison is relative humidity (RH). For this purpose the air temperature and specific humidity produced by ERA5 must be converted to RH. A similar exercise is performed for the radiosonde data, although radiosondes do measure RH directly. Is this because radiosonde BUFR data is used in this study? Wouldn't it be more practical to use the radiosonde's RH data directly? This would eliminate the need for a lengthy and complicated discussion of the conversion of dew point data to RH. Since one of the authors is directly affiliated to the manufacturer, this should be feasible. Furthermore, GRUAN data products include RH values.
A description of the GRUAN data processing for the M10 should be provided, as well as a discussion of the differences between GRUAN data processing and manufacturer data processing. Latter is essential for the reader to understand and interpret the observed differences in the measurement data.
The description of the plots in the figures often contain a listing of obvious and not necessarily relevant facts. Please focus on what the data presented in the plots tell us. For example, mean values of differences/biases are more interesting than the maximum values of outliers.
Throughout the manuscript the measurement uncertainties are referred to as % (which I presume to be relative errors) or %RH, which is quite confusing to the reader. Please be clear, and consistent, in this regard.
Section 3.3, that discusses the impact of changes to the data processing is not very convincing. Apparently changes to the processing software of the manufacturer data product are made regularly, and the data sets used for figure 4 and table 1 contain a mix of processing versions, where the time windows of the individual versions are not clear. I would prefer to see a comparison with two different processing versions applied to the complete period 2020-2024.

The numbers given in table 1 are hard to interpret. Are min,max and Q25-Q75 all necessary to make the point?
Similar applies to section 3.4, discussing the impact of switching from M10 to M20. The plots in figure 5 are quite cluttered, and it is difficult for the reader to follow the conclusions the authors draw from these plots. To what extend can real instrumental effects be separated from meteorological effects?

The difference between the daytime and nighttime results for Mime (upper two plots) are quite striking (the green and red traces seem to shift from aligning with the blue trace to aligning with the black trace). A qualitative discussion of this effect would be appropriate.
The authors should refrain from using subjective adjectives when describing results. E.g. write good agreement instead of a very good agreement, or even better: data A and B agree within x%.
Is the detailed discussion of RH over water or ice relevant? Radiosondes report RH over liquid water as stipulated by WMO.
Detailed comments

l20: Now the sentence implies that the data processing was developed by Meteo France. The correct situation is that the soundings are performed by Meteo France, but the processing is done with the software provided by the manufacturer. To prevent confusion refer to this stream as manufacturer data processing.

l21: the term GRUAN standard does not exist for data products. In case of the M10 radiosonde a GRUAN data product is under development. Better use the term GRUAN processing.

l23: 800 hPa -> 800 hPa, and

l23: is it 2% or 2%RH? This applies to various other locations in the manuscript as well

l25: maximum -> largest

l26: can you offer an explanation why the difference is larger for nighttime measurements?

l30: mention that 200-300 hPa corresponds to the altitude relevant for contrails

l31: indicate -> measure [or record]

l49: For completeness, you can mention that water vapor is important for the transport of latent heat.

l50: Thusfar you have described the role of WV in the atmosphere's energy budget, but did not discuss sources, whether natural or anthropogenic. Just state that aviation contributes to global warming by contrail formation in the upper troposphere.

l53: is calculating the right term here? Suggestion: deriving, obtaining, or establishing.

l54: However ... temperature. This stand-alone sentence doesn't add much.

l56: complex -> challenging

l63: reproducing -> capturing

l64: insert "in situ" after other hand.

l65: full -> global

l68-69: Awkward sentence. I assume you want to provide a reference to the dataset. Please rephrase.

l76-81: this sentence covers both homogenisation, trend detection and actual trends. Use separate sentences.

l81: "In the past" radiosonde intercomparisons is not a thing of the past, this method is still very much employed today. Insert the word performance before evaluations.

l85: this statement may be a bit too bold. I would prefer to say that it helped to improve data quality radiosoundings.

l88: insert comma after world.

l90: Goal of GRUAN is to provide reference quality measurements of ECVs, such as water vapor.

l94: capacitative sensors are used for RH measurements.

l101: Nothing spurious about time lag or radiative heating. Just state that important error sources for RH measurements are time lag at low temperatures and radiative dry bias.

l104: what are these "GRUAN standards"?

l106: what is the cause for this remaining bias between M10 and RS92? Calibration issues?

l108: are there other relevant studies on evaluating the performance of the M10?

l112: formed by -> constructed from GDPs of

l113: insert comma after radiosondes.

l123: what are these "GRUAN standards"?

l124: add: and to evaluate the manufacturer processing vs the GRUAN processing.

l133: including -> such as

l134: such as -> including

l150: p is not defined

l163: text suggestion: twice per day, at/around noon and midnight.

l167-169: there is an inconsistency in the numbers, unless the sampling frequency is 0.5Hz

l176: remove innovative (marketing jargon)

l185: calibrate -> correct (?)

l185: who is the manufacturer of the RH sensor, Meteomodem?

l187-190: marketing speak

l195: tunable heating system. Remove "which can be switched off"

l199: Which sensor is meant here (temperature, pressure, RH)?

l203ff: I assume this elaborate recalculation is necessary because you use BUFR data. If so, please mention this.

l217-219: in principle radiosondes measure/report RH over liquid water.

l230: more appropriate references on GRUAN are Seidel2009 (10.1175/2008bams2540.1) or Bodeker2016 (10.1175/bams-d-14-00072.1).

l231: the GRUAN data product for the M10 is still under development. The data processing does indeed exist.

l240: for above reason these data are not processed by GRUAN. Correct is to say that GRUAN processing method has been applied.

l255: I am not sure whether considering the mixed-phase is relevant here. WMO stipulates radiosondes to report RH over liquid water.

l261: I think you mean manufacturer-processing here.

l290: explain why should the location make a difference

l393: processings -> processing methods

l404: describe in the caption what is represented by the boxes and the lines.

l406: remove very

l408: precision -> uncertainty

l529-535: this is too speculative. I see no basis for this statement.

Reply

Citation: https://doi.org/10.5194/egusphere-2026-482-RC2
RC3: 'Comment on egusphere-2026-482', Anonymous Referee #3, 05 May 2026 reply

See attached file

Reply

Citation: https://doi.org/10.5194/egusphere-2026-482-RC3

Sidy Diarra, Nadège Montoux, Jean-Luc Baray, Philippe Keckhut, Jean-Charles Dupont, Antoine Farah, and Dunya Alraddawi

Viewed

Total article views: 584 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
376	170	38	584	51	45

HTML: 376
PDF: 170
XML: 38
Total: 584
BibTeX: 51
EndNote: 45

Views and downloads (calculated since 03 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	266	108	32	406
Apr 2026	107	60	6	173
May 2026	3	2	0	5

Cumulative views and downloads (calculated since 03 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	266	108	32	406
Apr 2026	107	60	6	173
May 2026	3	2	0	5

Viewed (geographical distribution)

Total article views: 617 (including HTML, PDF, and XML) Thereof 617 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 05 May 2026

Short summary

This study examined the quality of water vapor observations in the troposphere. We compared five years of balloon measurements over France with weather model reanalysis. The observations showed moister air and more frequent supersaturation than the model in the upper troposphere, where natural cirrus clouds and aircraft induced clouds formed. These results suggest that current models underestimate upper air humidity and need improvement to assess climate impacts of clouds.


Total:	0
HTML:	0
PDF:	0
XML:	0