Application of high-dimensional fuzzy <i>k</i>-means cluster analysis to CALIOP/CALIPSO version 4.1 cloud–aerosol discrimination

Zeng, Shan; Vaughan, Mark; Liu, Zhaoyan; Trepte, Charles; Kar, Jayanta; Omar, Ali; Winker, David; Lucker, Patricia; Hu, Yongxiang; Getzewich, Brian; Avery, Melody

doi:https://doi.org/10.5194/amt-12-2261-2019

Articles | Volume 12, issue 4

https://doi.org/10.5194/amt-12-2261-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

CALIPSO version 4 algorithms and data products

https://doi.org/10.5194/amt-12-2261-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 12, issue 4

Research article

|

12 Apr 2019

Research article |

| 12 Apr 2019

Application of high-dimensional fuzzy k-means cluster analysis to CALIOP/CALIPSO version 4.1 cloud–aerosol discrimination

Shan Zeng, Mark Vaughan, Zhaoyan Liu, Charles Trepte, Jayanta Kar, Ali Omar, David Winker, Patricia Lucker, Yongxiang Hu, Brian Getzewich, and Melody Avery

Download

Final revised paper (published on 12 Apr 2019)
Preprint (discussion started on 28 Jun 2018)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Review of Application of High-Dimensional Fuzzy K-means Cluster Analysis...', Anonymous Referee #1, 23 Jul 2018
- AC1: 'response to referee #1', Mark Vaughan, 07 Dec 2018
RC2: 'Review of : Application of High-Dimensional Fuzzy K-means Cluster Analysis to CALIOP/CALIPSO Version 4.1 Cloud-Aerosol Discrimination, by Zeng et al., 2018', Anonymous Referee #2, 28 Sep 2018
- AC2: 'response to referee #2', Mark Vaughan, 07 Dec 2018

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Mark Vaughan on behalf of the Authors (07 Dec 2018) Author's response Manuscript

ED: Referee Nomination & Report Request started (03 Jan 2019) by James Campbell

RR by Anonymous Referee #2 (23 Jan 2019)

RR by Anonymous Referee #3 (25 Jan 2019)

Suggestions for revision or reasons for rejection

Due to reviewer 1's inability to re-review the submission, I will be acting as reviewer 1.

There were significant changes to the manuscript and I thank the authors for putting in the time to make this a strong candidate for publication. There are only a few areas that could use a bit more clarification, and thus, I believe this study should be accepted with minor revisions. Such areas are mentioned below.

In page 5, 2nd paragraph where the COCA PDF development is explained: You mention the difference between the initial COCA PDF (4 dimensional) to the V3 COCA PDF (5 dimensional). A sentence about the differences in the V3 to V4 COCA PDFs is needed here since the results from CAD V3 to CAD V4 are compared later.

In Fig 4 there are two patches of elevated backscatter, one between 36.86 and 71.55 around 15 km and the other south of -69.6 around 12-15 km. Both are classified as aerosol using CAD V4, but cloud using the 3 class FKM. Maybe include a sentence on the discrepancy.

This also brings me to my next comment. The feature south of -69.6 is indicated as aerosol in the 2 class FKM, but cloud in the 3 class FKM. Presumably, going from a 2-class to 3-class is analogous to going from aerosol-cloud scheme to aerosol-water_cloud-ice_cloud scheme, and would likely be so in binary k-means clustering. In that case, a pixel identified as cloud, would fall within ice cloud or water cloud when going to a three class scheme. This figure shows that assumption is untrue, at least when using fuzzy logic. I think this phenomenon should be highlighted in the text, specifically as an example of the "twilight zone" between aerosol and diffuse high cloud.

Fig 7: for (c) and (f), I would choose to clip the scale at +/- 50%. Specifically, anything below 25% difference would not show on the plot as is, and, such a difference is not negligible. In fact, the authors have clipped the scale to +/- 10% when comparing zonal averages.

For the discussion on table 2, maybe highlight that the biggest difference seems to be FKM indicating features as aerosol that were determined as cloud using COCA. To me, this suggests the "twilight zone" again. Since the authors are suggesting that a global FKM CAD algorithm is likely less accurate when compared to current CAD, a bit more discussion on the impacts of these features should be included somewhere.

Figure 12 (b) is very difficult to read due to the very similar color choices. since 7 variables are being plotted, I'm not sure how much can be done, but it is nonetheless hard to analyze, especially when printed.

Hide

ED: Publish subject to minor revisions (review by editor) (22 Feb 2019) by James Campbell

AR by Mark Vaughan on behalf of the Authors (26 Feb 2019) Author's response Manuscript

ED: Publish as is (19 Mar 2019) by James Campbell

AR by Mark Vaughan on behalf of the Authors (27 Mar 2019) Manuscript

Short summary

We use a fuzzy k-means (FKM) classifier to assess the ability of the CALIPSO cloud–aerosol discrimination (CAD) algorithm to correctly distinguish between clouds and aerosols detected in the CALIPSO lidar backscatter signals. FKM is an unsupervised learning algorithm, so the classifications it derives are wholly independent from those reported by the CAD scheme. For a full month of measurements, the two techniques agree in ~ 95 % of all cases, providing strong evidence for CAD correctness.