the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Evaluation of EarthCARE retrievals of ice microphysics and vertical wind using in-situ aircraft observations
Abstract. The novel observational capabilities of the EarthCARE satellite promise the most accurate retrievals of the vertical profile of ice clouds yet achieved from space, but independent evaluation is essential. In this paper we use in-situ sampling from five underflights of EarthCARE in ice clouds between -10 and -45 °C during the UK "VERIFY" and Canadian "ECALOT" campaigns to evaluate and improve its microphysical and vertical-wind retrievals, as well as testing prior assumptions such as the mass–size relationship and the radar ice scattering model. We find that, to a good approximation, the small-scale fluctuations in radar-measured Doppler velocity can be attributed to vertical wind and the larger-scale averages to ice terminal fall speed. When EarthCARE’s "C-CD" algorithm is updated to exploit this finding, its vertical wind retrieval is able to capture the amplitude and phase of gravity waves measured by the aircraft. Retrievals of IWC and extinction by the radar-only "C-CLD" algorithm, the lidar-only "A-EBD" algorithm and the synergistic "ACM-CAP" algorithm, are in good agreement with the aircraft but highlight the important challenge of retrieving ice particle density. Comparing the 94-GHz radar reflectivity observed by EarthCARE with values computed from the aircraft probes enables us to successfully validate the Self-Similar Rayleigh-Gans scattering model even when non-Rayleigh scattering reduces the reflectivity below the equivalent Rayleigh value by 15 dB. However, the calculations of reflectivity-weighted terminal fall speed from the aircraft are systematically 15–30 % higher than measured by EarthCARE’s Doppler radar, suggesting the need for further work on models of ice fall speed.
Competing interests: PK is a member of the editorial board of Atmospheric Measurement Techniques. RH and DD serves as editor for the special issue.
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.- Preprint
(2536 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-5054', Alain Protat, 29 Sep 2026
-
AC3: 'Reply on RC1', Robin Hogan, 03 Nov 2026
We thank the reviewer for his constructive review.
You have a point about the figures, but your suggestion would separate all the variables for a particular case into different figures, then making it difficult to compare different variables for the same case. We propose a compromise: in the revised version we will change the order of the figures so that what are now Figs. 1, 3, 5, 7 and 9 (i.e. the radar variables) will be presented first, followed by what are now Figs. 2, 4, 6, 9 and 10 (i.e. the retrievals).
A good suggestion on scatterplots: the new version adds a multi-panel figure containing scatterplots comparing the various variables between aircraft and EarthCARE, and it does indeed help understanding the errors in the retrievals. However, since the quantitative comparison in Table 2 considers only datapoints taken within 30 km, or up to 5 minutes apart, and the scatterplots do the same, there is no significant degradation of agreement with distance, so it did not seem necessary to plot the difference as a function of separation distance or time.
This colour scale we use is already widely used for quicklooks using EarthCARE data. One of the authors has the most common red-green colour-blindness, but he prefers this colour scale to the alternatives for EarthCARE plots because the plots span multiple orders of magnitude and allow differences to be distinguished visually in all parts of the scale. By contrast, colour scales that provide a strictly dark-to-light scale applied to a plot spanning multiple orders of magnitude typically do not allow differences of even as much as a factor-of-two to be easily distinguished visually.
Citation: https://doi.org/10.5194/egusphere-2026-5054-AC3
-
AC3: 'Reply on RC1', Robin Hogan, 03 Nov 2026
-
RC2: 'Comment on egusphere-2026-5054', Anonymous Referee #2, 30 Sep 2026
The authors evaluate in this manuscript recent EarthCARE retrieval products of ice microphysics and vertical wind using aircraft in-situ data from five underflights collected during different aircraft campaigns.
Overall, I find the paper very well written and an important study for the evaluation of the new EarthCARE retrieval products. It really highlights the potential and quality of the new global satellite dataset. I have a few comments, suggestions, and questions which I would like the authors to address. I think the paper needs some revision before it can be published.
Major Comments:
One of my major concerns regarding the analysis is that uncertainties of the airborne observations and EarthCARE retrievals are not properly discussed. From many previous publications and conference presentations I got the impression that airborne in-situ data (for example IWC) from different probes measuring the same cloud can quite significantly deviate from each other. Considering the multitude of previous aircraft campaigns, are there no uncertainty estimates available for these basic quantities like IWC, PSD, or vertical air velocity? I understand that uncertainty estimates might be difficult to derive from such complex probes but as a reader, I would at least like to know a rough number. The only uncertainty value I found was the 50% IWC uncertainty for particle “rebound” (L. 75-77). While 50% is already quite big in my opinion, I wonder what is the total uncertainty in IWC? And how much more uncertain are more “difficult” quantities, such as PSD? Similar arguments also apply to the EarthCARE retrievals. How big are the uncertainties and how do they compare to the differences between EarthCARE and aircraft data (or forward simulated radar quantities)? I assume the ESA retrieval products do contain some uncertainty estimates, so why are they not discussed or shown?
Many acronyms for retrieval products and instruments are used in this study. They are properly introduced and explained but it might be convenient for the reader to have a list of acronyms and abbreviations provided at the end.
Specific comments and typos:
L. 32: “radar reflectivity at the level of the aircraft” What is the blind range of the airborne radar? I guess it is not zero, hence the first useable range gate will not be exactly at flight level.
L.32-33: Please specify what is meant with “significant amounts of supercooled water”. What threshold was used? At L. 55 you state that cases with any SLW were excluded. What is the sensitivity level of the airborne SLW probes? It should also be noted that absence of SLW at flight level does not guarantee SLW-free conditions elsewhere in the cloud. Wouldn’t it be possible to use the PIA of the airborne W-Band radar as indicator of SLW? Or was the aircraft equipped with a microwave radiometer?
L. 36-37: I am a bit confused: You mention that Z94 is sensitive to mass-size relation and scattering assumptions. I completely agree. But I don’t understand how you can distinguish the effect of each of them by comparing measured and forward simulated Z94? You have two unknows and only one observable.
L. 69: A reference for the anti-shattering algorithm should be provided.
L. 73-74: Why exactly 50%? A reference for this value should be provided. As you mention in the next sentence, the grey band in the figures show this potential 50% uncertainty. But how accurately is this known? Can’t the rebounding error be dependent on the particle properties you are probing. What other error sources or uncertainties are known for the Nevzorow probe?
L. 89: Couldn’t the m-D relationship also be strongly altered by the ice particles having simply a different shape (e.g. columns vs plates)? Does the density factor in the end account for shape and riming effects?
L. 95-100: Why is A(D) not dependent on fp? Increasing riming will also have an effect on the cross sectional area, or?
L. 102: Is Dm simply the mass median diameter?
L. 109: What is meant with “all radar frequencies”? Aren’t you only simulating W-band in this manuscript?
L. 115-117: Shouldn’t the SSRGA coefficients also depend on the density factor df (see for example study of Maherndl et al., QJ, 2023, https://doi.org/10.1002/qj.4573)?
L. 133: Please specifiy what “some horizontal averaging” means? Is the horizontal averaging window adjustable? In which range?
L. 139: What is meant by “geophysical variability”? Changes in particle properties and vertical air motion? Please specify.
L. 145: Consider including a reference to the section you are referring to.
L. 198: “how whether”
L. 369: Did the authors consider to test other hydrodynamic models, such as Böhm 1992? For example, in Karrer et al., JAMES, 2020 (their Fig. A2, https://doi.org/10.1029/2020MS002066 ) the Heymsfield-Westbrook-2010 model shows the highest estimates for sedimentation velocities of synthetic particles (especially for larger sizes which will dominate reflectivity weighted Doppler velocities).
Citation: https://doi.org/10.5194/egusphere-2026-5054-RC2 -
AC1: 'Reply on RC2', Robin Hogan, 03 Nov 2026
We thank the reviewer for their constructive review.
This is a good point about errors. We have updated the revised manuscript as follows:
- For aircraft measurements of vertical wind we cite Mallaun et al. (AMT 2015) who reported an uncertainty of 0.2 m/s using the Rosemount sensor used in ECALOT.
- For the particle size distribution (PSD), assuming the shattering fragments have been removed by the interarrival-time algorithms applied to both aircraft datasets, one could estimate uncertainties in bin counts due to Poisson counting statistics, but ultimately the errors in moments of the size distribution are dominated by assumptions on the nature of ice particles. The resulting errors are difficult to estimate, so it is standard practice to characterise them by comparing to independent measurements of the same quantity. Even then there are uncertainties. In the case of ice water content, the error is dominated by the uncertain mass-size relationship. We have added a scatter plot to the revised paper comparing IWC derived from the optical-array probes and the Nevzorov probe, which shows a fractional root-mean-squared difference of 20% in 5-km averaged quantities. However, as the reviewer noticed, the usual data processing practice differs between datasets from the FAAM and Convair aircraft, with the IWC from the Convair's Nevzorov probe usually scaled up by 50% due to concerns about rebounding, even though this is not done with FAAM data. The 20% agreement above is obtained without rescaling of either dataset. Furthermore, the comparison of radar reflectivity factor in the first four Rayleigh-dominated cases do not support a scaling-up of the Nevzorov probe IWC by 50%, since this would imply a scaling of the particle mass by the same amount, which would increase the aircraft-simulated reflectivity factor by 3.5 dB leading to much poorer agreement. Clearly aircraft data are not perfect, but they are the best we have in terms of evaluating EarthCARE products, and we hope that the revised discussion adequately discusses their remaining uncertainties and the various lines of evidence we can collect to evaluate each product.
- The retrievals themselves do indeed provide error estimates, typically based on propagating observational errors through the retrieval scheme and where possible accounting for "model error", i.e. uncertainties in the assumptions that are made in the retrieval. However, ultimately these error estimates themselves need to be validated by comparisons such as in this paper. We now compare the reported errors in the retrievals with the differences to the aircraft data. Accepting that the aircraft data are not perfect and there is additionally some co-location error, we find that the reported errors in A-EBD particle extinction are reasonable for the first two cases (VERIFY6 and VERIFY2), but for the third (ECALOT3) where the lidar was at the limit of its sensitivity, the product significantly overestimated the error. By contrast, the errors in the ACM-CAP retrievals of extinction, IWC and effective radius are significantly underestimated. These findings will be valuable in future revisions to these algorithms to improve the error estimates that come with them.
We believe that the number of acronyms is fairly typical of papers in AMT, and therefore that a glossary is not really needed, especially given that it is easy to search in an electronic copy of the paper for the meaning of an acronym. We rather see lists of symbols to be useful in highly mathematical papers (unlike this one) where it is difficult to search for where a single-letter symbol is defined in the paper.
Responses to specific comments:
L32: There is no aircraft radar: we are referring to the EarthCARE radar reflectivity at the level of the aircraft, which is now stated. The "aircraft radar reflectivity" is calculated from the aircraft size distribution using Eq. 8.
L32-33: We used the Nevzorov liquid probe to detect significant supercooled liquid, but there were not enough cases for it to be worth using formal exclusion criteria. One such excluded flight was ECALOT7 - see Fig. 8 of Qu et al (2026) where the LWC measurements coincide with EarthCARE retrievals of supercooled liquid. Certainly there may be supercooled water elsewhere in the cloud, but it is only if it occurs at the flight level that it invalidates the comparison between EarthCARE and the aircraft measurements. The PIA cannot be used to infer supercooled water because there was invariably lower level liquid cloud in the profile and ECALOT took place over land so PIA was not reliable.
L36-37: Yes agreed, the comparison evaluates the overall simulated reflectivity so from an individual case it is not typically possible to evaluate the strength of the two effects separately. This is now stated. However, several of the cases have particles that are too small to scatter outside the Rayleigh regime so they evaluate only the mass-size relationship. The conclusions does state that more cases are needed in non-Rayleigh conditions to provide further confidence in this part of the comparison.
L69: A reference to Korolev and Field (2015) has been added.
L73-74: The 50% is based on unpublished work by one of the authors (Korolev) and is typically applied in processed Nevzorov data provided from the Convair aircraft. However, it is not applied to the processing of the same instrument on the FAAM aircraft, and if we were to increase the mass-size relationships to match it, the aircraft-simulated reflectivity in Rayleigh-scattering conditions would increase by 3.5 dB and be in poorer agreement EarthCARE radar measurements. Therefore we have altered the text to say that there is some uncertainty in the Nevzorov probe due to possible rebounding, but no longer show the light grey shaded band on the IWC comparison in Fig. 6f and similar figures. In response to the reviewers more general comment about errors, we also show various scatterplots including comparing IWC between the Nevzorov probe and the optical array probes with the mass-size relationships described in the paper, and find that the fractional root-mean-squared difference is only 20%.
L89: Indeed, rapid depositional growth can lead to higher density sector plates, for example in the case shown in Fig. 3 of Hogan et al. (2006). This is now stated.
L95-100: This is indeed a source of uncertainty, which we now state, but we don't have a reference for any change to this relationship with density factor. We now mention this as a potential contributor to the difference in extinction in the final two cases where the density factor was slightly larger than 0.
L102: Dm was defined in the text just before Eq. 7, now it is defined in its own Eq. 7. It can be thought of as a "quasi-mass-weighted" mean diameter, and it follows the definition of Field et al. (2005).
L109: We are indeed simulating only 94 GHz, but it doesn't hurt to say that this value is frequency independent.
L115-117: More complex formulations are possible but here we test the assumption used currently in the ACM-CAP algorithm and in the final case with significant non-Rayleigh scattering, it appears to work, but we have now cited Maherndl et al. (2023) in the conclusion where the need for further EarthCARE/aircraft comparisons in non-Rayleigh conditions is highlighted.
L133: We updated the text in the revised version to state that averaging is performed on a scale for 4-km horizontally and 500-m vertically.
L139: Yes, variations in fall speed - this is now stated.
L145: This is refering to the revised C-CD algorithm, not the analysis of the present paper, as is now clarified.
L198: Fixed.
L369: We have added a reference to Karrer et al. (2020) in the conclusion, but their Fig. A2b rather implies that for the aggregates/irregular particles that dominated the cloud sampled in our paper, the Heymsfield & Westbrook model was in the middle of the three models considered.
Citation: https://doi.org/10.5194/egusphere-2026-5054-AC1
-
AC1: 'Reply on RC2', Robin Hogan, 03 Nov 2026
-
RC3: 'Comment on egusphere-2026-5054', Anonymous Referee #3, 05 Oct 2026
This manuscript describes validation of EarthCARE Level 2 products of ice microphysics and vertical wind using 5 flights installing in-situ cloud sensors.
EarthCARE/CPR is first sensor that can observe vertical Doppler information of clouds from a satellite. Data processing algorithms to estimate ice cloud microphysics and vertical air-velocity has been developed using Doppler information, so far. However, as to evaluate those accuracy, in-situ observations using airplane under satellite is necessary. This paper shows not only accuracy of those products but also difference between estimated cloud microphysics and in-situ measurement. I think this paper is very valuable for algorithm developers and data users, because it shows several problems of current algorithms and useful suggestions for new algorithm developments.
I judge that this paper is acceptable for publishing as it is. However, I will give some minor comments for your reference.
1. EarthCARE data products and processing algorithms are not familiar to almost all readers. It is useful to show the table of level 2 data processing (C-CD, C-CLD, A-EBD, ACM-CAP) including its products and its release version used in this paper. I think data processor version is very important because results may change with algorithm version. Please add some explanation about version such that first letter shows major version and second letter shows minor version on “BD-baseline version” “BA data product baseline”
2. Latitude is used for horizontal axis in all figures. But I think it is more comprehensible to use distance from satellite passing. I also want to know rough time difference between satellite and airplane observation.
3. Doppler velocity and estimated vertical velocity in VERYFY2 and ECALOT6 show very good agreement to airplane measurement and show gravity-wave-like perturbation. I am surprised by these agreements because I thought the vertical velocity measurements from the airplane were difficult and less accurate. Are there any inferences on cause of gravity waves? Mountain topography?
Citation: https://doi.org/10.5194/egusphere-2026-5054-RC3 -
AC2: 'Reply on RC3', Robin Hogan, 03 Nov 2026
We thank the reviewer for their constructive review.
1. We had mentioned the baseline for most of the products but now do so for all of them, and clarify the major/minor meaning of the letters. We don't see the value of a table just to state the baseline when the information is already available in the text.
2. The use of latitude is consistent with many other papers on EarthCARE, e.g. Qu et al. (2026). We now clarify in the context of referring to the Table 2 that for a nominal aircraft ground speed of 100 m/s, the requirement for the EarthCARE/aircraft comparisons in the Table to be within 30 km means that the clouds are typically sampled within 5 mins of each other. Additional dotted lines are now presented on the radar reflectivity figures (e.g. Fig. 1a) to indicate the range within 30 km of the overpass, and to give the reader an idea of horizontal scale.
3. Mallaun et al. (2015) reported an accuracy of 0.2 m/s for the Rosemount sensor that was used in ECALOT - this is now stated. VERIFY2 took place over the Atlantic to the north-west of Scotland in a southerly warm-sector flow. Satellite imagery from NOAA-21 showed very extensive wave clouds over Scotland, Wales, Ireland and downstream over the sea downstream of them, so indeed the gravity wave would appear to be orographically triggered. ECALOT 6 took place in a westerly warm-sector flow and some limited wave features could be discerned in satellite imagery but much less extensive than VERIFY2. We have now made these points in the text and added a figure showing the NOAA-21 imagery, with EarthCARE's narrower MSI swath superimposed.
Citation: https://doi.org/10.5194/egusphere-2026-5054-AC2
-
AC2: 'Reply on RC3', Robin Hogan, 03 Nov 2026
Data sets
Colocated EarthCARE retrievals and aircraft in-situ measurements of ice clouds (Version 1.0) R. J. Hogan et al. https://doi.org/10.5281/zenodo.22043988
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 578 | 347 | 70 | 995 | 70 | 64 |
- HTML: 578
- PDF: 347
- XML: 70
- Total: 995
- BibTeX: 70
- EndNote: 64
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
See attached