| 15 Oct 2026
Evaluation of EarthCARE’s Scene Construction Algorithm Using Airborne Observations
Abstract. This study evaluates the performance of EarthCARE’s scene construction algorithm (SCA) using airborne observations collected by the National Research Council Canada’s Convair-580 research aircraft during the EarthCARE Commissioning Calibration/Validation Campaign in Ottawa (ECALOT). EarthCARE’s three-dimensional (3D) SCA employs a radiance-matching technique and nadir retrievals to construct 3D atmosphere-surface conditions around the nadir cross-section. While the SCA can be applied across the ~150 km swath of EarthCARE’s Multispectral Imager (MSI), its operational application is limited to a ~30 km-wide region centered on the nadir track., which encompasses the 5 km across-track by 21 km along-track assessment domains and their associated buffer regions. The SCA facilitates lateral transport of photons in Monte Carlo radiative transfer simulations, and so its results influence simulated radiative quantities that are used, in part, to perform radiative closure analyses. Two complementary evaluation strategies are applied. First, the Convair-580 underflew EarthCARE on a path angled at 20° to the satellite’s track. Off-nadir measurements of cloud microphysical variables made in situ by the Convair-580 are used to assess SCA predictions for shallow cumulus and stratocumulus clouds. Second, for nimbostratus clouds, the Convair-580’s side-viewing W-band radar reflectivities are compared to horizontal planes of reflectivities based on SCA allocations of nadir reflectivities from both Convair-580 and EarthCARE’s Cloud Profiling Radar. Evaluation against in situ observations showed little variation in mean bias error (MBE) and root mean square error (RMSE) for both liquid water content and effective radius at distances of up to ~12 km from the nadir track, indicating consistent SCA performance within this range. Comparison with side-viewing W-band radar reflectivity also yielded low MBE and RMSE, with correlation coefficients of 0.88 and 0.80 for Flights 6 and 5, respectively. The newer version of the SCA, Baseline BC, generally outperformed the earlier Baseline BA, particularly for the more complex cloud scene observed during Flight 5 on January 27, 2024.
Review of a paper by Qu et al. entitled “Evaluation of the EarthCARE Scene Construction Algorithm Using Airborne Observations,” .
The authors used a high-resolution numerical weather prediction model and data from the ECALOT field campaign to evaluate EarthCARE’s scene construction algorithm. The study focuses on evaluating two products, baselines BA and BC. Baseline BC is an improved version of BA that extends the search for a donor pixel from 200 km to 400 km and removes the solar zenith and azimuth angle restrictions.
In the first part, the authors evaluate the liquid water content and particle size derived from the scene construction algorithm using a modeled water-cloud field. Unlike the aircraft data, this approach excludes the effects of retrieval errors and temporal and spatial sampling mismatches. The agreement in liquid water content and particle size is good, and the bias is nearly constant with distance from the nadir track. TOA radiances at 0.67 and 10.08 microns are also compared and found to be in good agreement.
In the second part, the authors evaluate data from 12 flights conducted during the ECALOT campaigns, including a water-cloud case and ice- and mixed-phase-cloud cases. The results show that the scene construction algorithm generally performs well. There are slight improvements from the BA algorithm to the BC algorithm. When scenes are complex, the improvements are more pronounced. The constructed liquid water content and radar reflectivity agree well with observations. The RMS difference increases with distance from nadir, especially for liquid water content when the distance is greater than 10 km. The comparison as a function of distance from nadir is limited to 10–14 km because of observational limitations related to the flights and the horizontally looking radar.
The manuscript is well written. I have only minor comments and requests for clarification.
1) In the discussion of the results, the findings from Part 1 (the modeling results) and Part 2 (the aircraft observations) are not linked very well. In fact, Part 1 shows that the bias is constant with distance from the nadir track, whereas the results from Part 2 show that the bias increases with distance (e.g., Figure 7 versus Figure 14). I would like to see some discussion explaining and reconciling this difference.
2) Is the 0.67-micrometer channel used in the scene construction algorithm? If so, should we expect only small biases when the 0.67-micrometer radiances are compared?
3) Line 203: For the swath-sampling case, did you exclude the recipient pixel when searching for a donor pixel?
4) Figure 11b–e: Do the colors have any meaning? If so, could you please provide a color bar? The same comment applies to Figure 13b and c.
5) Figure 12: Does the color indicate population density? Please explain this in the caption.
6) Line 430: I am not sure that I understand the meaning of this sentence. Why does the extensive overcast condition limit the ability? Is it because the clouds are uniform?
7) Line 491: The statement that the BC result is better than the BA result needs further explanation. If I understand correctly, BA searches within 200 km along track, whereas BC searches within 400 km. In addition, the SZA and SAA restrictions are removed for BC. It is counterintuitive that using less restrictive criteria leads to a smaller error. Is this because failing to find a match in BA leads to the use of a donor pixel with a larger difference? Also, is finding a donor pixel at a greater distance better than finding a closer donor pixel with a larger difference?