Algorithm for Improved Stereoscopic Cloud-Top Height Retrieval Based on Visible and Infrared Bands for Himawari-8 and FY-4A
Round 1
Reviewer 1 Report
The manuscript by Lee and Shin presents an interesting algorithm to improve the stereoscopic CTH retrievals based on FY-4A and Hamawari-8 satellites. The algorithm and its performance are well presented, and is helpful for better understanding cloud properties. The article is well organized and discussed. However, there are still some concerns on the work, and some further discussions are suggested to improved the paper.
- There have been a couple of previous studies on the evaluation of AHI or AGRI CTH results. They would be helpful to better demonstrate the performance of the new algorithms,and should be cited and discussed (https://doi.org/10.1175/JTECH-D-18-0231.1 and https://doi.org/10.1007/s00376-021-0337-2).
- Although the prototype dual-GEO CTH algorithm is now completely new, a schematic figure may be helpful for readers to better understand this work.
- One of the advantages of the new algorithm is the improvement on multi-layer clouds. It is noticed that the multi-layer clouds with a upper ice and lower water clouds can be well detected using only Himawari-8 (Wang et al., 2019, doi:10.1016/j.rse.2019.02.024), and even their optical and microphysical properties of both layers may be detected (Teng et al., 2020, 10.1029/2020GL088941), would such information helpful for further improvement on their CTH retrievals? This would be really interesting, and is suggested to be discussed in the revision.
- Some physical explanations on the reasons of the new algorithm improvements should be added.
- The comparisons between active CTH and this study in Figs. 5d, 7d and 9d are difficult to read due to the reflectance background, and the performance of new algorithm can hardly be illustrated. Thus, the figure has to be improved.
- Besides the comparison with active datasets, a pixel-to-pixel comparison between the new results and the AHI operational CTH products is suggested, e.g., the CTH distributions within the three regions of the three studied cases.
- It is noticed that there are still biases over 2km, would such biases reasonable for the dual-GEO CTH algorithm?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper describes a novel stereoscopic method for retrieving cloud top height (CTP) of thin upper-level cloud as well as lower-level cloud using a combination of VIS and IR channels on geostationary satellites, Himawari-8 and Feng Yun-4A. The results are compared with those from CALIOP and CPR. The work is an advance from Lee et al. (2020). Although the paper is quite long, the repeated material from earlier papers is concisely presented. The use of image cloning to insert the IR patch into the VIS image is very appropriate, yielding promising results.
The results are good, but as expected there is some difficulty with very thin (IR) high cloud. It would be useful to identify a sensitivity threshold for this feature. It would also be good to give more detail of the nature of the secondary modes in the error PDFs; ie actual lower layers or between layers.
The paper is very well-written and the analysis is systematic.
Specific comments
l112. The full specifications of the Himawari8 and FY-4A instruments can be referenced rather than listed in Table 1. Indeed only the channels used in the algorithm need to be discussed.
L166. Cross-correlation is a well-understood statistic. It is not clear why this feature of the original algorithm is detailed. It is also detailed in the earlier paper.
L269. “This approach is based on the fact that the gradients of the inserted image in the target image are closest to those of the source image.” I’m not sure that this is the best summary description of image cloning: perhaps something like ‘Image cloning aims to minimise the difference in gradients between the source and cloned images while matching the target image at the source boundaries’. (The method is better explained in line 327.)
L286. The caption of figure 4 should clarify that the target image is the original image, rather than the cloned image.
L342-372. The discussion of figure 5 should include the region from 35-40S, where the new method seems to have some difficulty with the high (thin) cloud.
L407-426. Is it really necessary to define MAE, MBE and RMSE in equations?
L441-444. The inference that the new algorithm has difficulty with very thin upper (which is not surprising) could also be drawn from figure 5.
l447-453. The error distributions in figure 8 are strongly bimodal. It should be determined whether the secondary peak is at the CTH of the lower cloud or somewhere in between. I think it is the latter.
There should also be a comment on Himawari8. It has some well-known problems; eg Huang et al. (2019, JAOT).
L483. Each example shows the difficulty with optically thin high cloud. Can you quantify this feature in terms of a threshold?
L493-506. As before, it would be useful to specify the detailed nature of the secondary modes in the error PDFs of figure 10; ie lower clouds or between layers or both.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have revised the manicurist according my comments, and it could be accepted for publication in remote sensing.
