| 06 Aug 2026
EarthCARE Doppler Velocity validation using ACTRIS ground-based cloud radar network
Abstract. The Earth Cloud, Aerosol, and Radiation Explorer (EarthCARE), launched in May 2024 by ESA and JAXA, carries the first spaceborne Doppler cloud profiling radar (CPR) capable of providing global observations of hydrometeor vertical motion. The retrieval of Doppler velocities from space is technically challenging due to satellite motion–induced spectral broadening, non-uniform beam filling, low signal-to-noise ratios, and antenna mispointing. In this study, we evaluate EarthCARE CPR Level-1b (L1b) and Level-2a (L2a) uncorrected and best estimate Doppler velocity products during the first 18 months of the mission (June 2024–November 2025) using ground-based cloud radar observations from the European Aerosol, Cloud, and Trace Gase Research Infrastructure (ACTRIS) network. Ground-based measurements are forward-simulated with an orbital radar simulator to ensure geometric and sampling consistency with the CPR, enabling a harmonised, profile-based comparison across multiple sites in Europe and the polar regions.
After applying a harmonised preprocessing and filtering procedure to ensure dataset consistency, systematic Doppler velocity biases are quantified for individual overpasses and statistically across sites. The uncorrected L1b and L2a Doppler products exhibit biases that correlate with the ESA antenna mispointing estimate, confirming the dominant contribution of this effect. In contrast, the L2a best-estimate Doppler velocity product effectively mitigates these biases. For sites with sufficient overpass sampling for robust statistics, including Ny-Ålesund, Lindenberg, Jülich, and Neumayer Station, the median residual bias of the best-estimate product ranges from −0.05 ms-1 to 0.01 ms-1, well within the mission requirement of 0.1 ms-1. Sensitivity analyses indicate only a weak dependence of the retrieved bias on spatial collocation distance (≤ 100 km) and temporal averaging window (±30 min).
These results demonstrate the robustness of the EarthCARE CPR Doppler velocity retrieval following application of the L2a correction scheme and confirm the suitability of the ACTRIS cloud radar network for long-term validation. The validated performance of the L2a best-estimate product provides confidence in the use of EarthCARE data for global investigations of cloud dynamics and hydrometeor vertical velocity climatologies.
This manuscript describes Doppler product's validation of EarthCARE Cloud Profiling Radar (CPR) using ACTRIS ground-based radar.
My review comments are as follows.
Major comment
1.
For CPR Doppler validation, we should consider two errors: random error and bias error. This manuscript focused on bias errors caused by CPR antenna mispointing, but its random errors of CPR and ACTRIS are not mentioned. That is why error profiles are not mentioned and only fitted bias errors are used after section 4. This is no problem, if you mentioned clearly that the main aim is validation of C-CD antenna mispointing bias. If you aim to general Doppler validation (both random and bias) using ground radar, more analysis about evaluation of random error should be added.
2.
On the validation of L2 C-CD's antenna mispointing bias, Kim et al. (2025) already showed similar analysis and showed its accuracy using NSA site (not ACTLIS) and Nemayer site (ACTLIS). You should mention Kim's results in introduction.
3.
We expected L2 Doppler accuracy is better than L1 Doppler. Comparing vL1 and vL2 in Figure 3,4,5 and Table 3, vL2 random error seems almost the same or a little bit worse compared to vL1. I guess that some vL2 include some data processing besides antenna mispointing correction and it gives worse effect on Doppler estimation. If so, this becomes interesting issue for this paper.
4.
There is no description why you chose 7 ACTRIS sites in this study. From Table 1 of Feuillard et al. (2026), ACTRIS also include W band radar in Jülich, Palaiseau, Lindenberg, Leipzig. These explanations are needed.
5.
L247 "The sites of Ny-Ålesund, Lindenberg, Jülich, and Neumayer have a sufficient amount of matching satellite and ground-based radar data available to analyses the Doppler velocity, while the sites of Bucharest, Granada, and Potenza have too few overpasses for a robust comparison (see Table 2)."
I think that the combined number of Lindenbereg (15+5) is almost the same to combined number of Granada(17) in Table 2. You should not omit Granada result, if you include Lindenbereg result.
6.
For evaluation of antenna mispointing bias in Figure 4, authors use median value (P50) instead of mean value. I think it is not common for statistical evaluation. Please explain this.
7.
Figure 3 and 5 only shows the results of Ny-Ålesund, Jülich,and Potenza. However, if possible, it is better to show results of the other sites (Lindenberg, Bucharest, Granada and Neumayer) with similar figures to understand reliability of analysis.
Minor comment
1.
On the L1 and L2 processing versions, author using DA baseline, CA to CD baseline, BA baseline without detail explanation. Author should explain that the data processing software of CPR L1 & L2 has been changed several times, and CPR data products have different values for improving algorithms.
author should explain that DA baseline means that major version is D and minor version is A and version name updates with alphabetic order. It is better to add the table of versions of L1 and L2 with release time and some explanation.
2.
In the Figure 1 explanation, "during this study are displayed in blue" should red
3.
L94, we could not understand data availability of descending orbit in MAAP. JAXA G-portal site also provides EarthCARE product including ESA L2 product including descending orbit. You may use those data for your analysis if you need.
4.
L96, "We traced the issue down to incorrect geo-locations in the metadata, which occurred only in the EarthCARE's CPR data. These issues have been fixed in the latest baselines released after October 14, 2025. Older data will be corrected during the next reprocessing of the entire EarthCARE data set."
Does "the latest baselines released after October 14, 2025" mean L1 or L2 version? I never heard of incorrect geo-location only in CPR. Could you show reference of this?
5.
Figure 3 & 5, I find some vL2a dot (yellow) or vbest (blue) is not shown in the figure. I think it exists out of range from 0.75 to -1.25. To evaluate three values, almost all data points should be shown. Those graph legends overlapping the graph should be placed outside of frame. If CPR Doppler velocity has no random error, red and yellow dot should appear on blue line, and blue cross should appear on zero line. To show them clearly, I recommend adding separate graphs that include only blue cross.
6.
In Figure 4 (b) (c), whiskers that shown 1.5-98.5 percentile line are missing at Potenaza and Granada. Please draw them. However, I also think that the box showing 25-75 percentile is enough to show here.
7.
In Table 1, the acronym of Doppler velocity pointing bias use L2aMPB,but to show velocity bias, I propose to use VL2AMPB (A: capital).
8.
L165 satellite-equivalent Doppler velocity field (vsat). But it makes confusion using "sat" subscript for the ground radar data of ACTRIS. Please modify it.
9.
L152 "Variations in PRF affect both the radar’s vertical and Doppler velocity resolutions and were therefore accounted for during preprocessing."
PRF is not related to the vertical resolution or Doppler velocity resolutions, but it is related to Doppler accuracy.
10.
L72 "The two products differ in their vertical integration along-track: 0.5 km for CNOM and 1.0 km for C-CD."
vertical -> horizontal
11.
L261,"In contrast, the uncorrected CPR Doppler velocity products, VL1–vsat (red) and VL2a–vsat (yellow) exhibit substantially larger biases resulting from the uncorrected non-uniform beam filling (NUBF) and Doppler spectrum broadening and therefore correlate well with the antenna pointing bias (blue line)."
I cannot catch this sentence. Uncorrected non-uniform beam filling (NUBF) and Doppler spectrum broadening is relating to random errors and not relating to bias errors. That is why they are scattering around the antenna mispointing bias.
12.
L278 "When biases of the uncorrected products are comparable, the spread of the VL2 data is typically wider. This behavior is also observed for the three sites with low data availability."
I cannot catch these sentences. Do you mean that the antenna mispointing estimation in C-CD is unreliable around three sites (Bucharest, Granada, and Potenza)?
13.
L289 "The validation of CPR Doppler velocities shows no significant dependence on the site-to-overpass distance within a radius of 100 km (not shown)."
I think that it is useful to show this result like Table 3.
14.
L324 "The results demonstrate that, although the uncorrected CPR Doppler velocities are strongly affected by antenna mispointing, non-uniform beam filling, and Doppler spectrum broadening, these effects are effectively mitigated in the ESA L2a best-estimate Doppler velocity product."
This paper only shows the bias error caused by antenna mispointing is reduced by L2a processing, but not show random error reduction caused by the non-uniform beam filling, and Doppler spectrum broadening.
15.
L328 "which lies well within EarthCARE’s mission requirement of 0.1 ms−1"
I firstly hear that EarthCARE’s mission requirement is 0.1 ms−1. Please show the evidence.