Exploration of machine learning methods for the classification of infrared limb spectra of polar stratospheric clouds

Sedona, Rocco; Hoffmann, Lars; Spang, Reinhold; Cavallaro, Gabriele; Griessbach, Sabine; Höpfner, Michael; Book, Matthias; Riedel, Morris

doi:https://doi.org/10.5194/amt-13-3661-2020

Articles | Volume 13, issue 7

https://doi.org/10.5194/amt-13-3661-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/amt-13-3661-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 13, issue 7

Research article

|

08 Jul 2020

Research article |

| 08 Jul 2020

Exploration of machine learning methods for the classification of infrared limb spectra of polar stratospheric clouds

Rocco Sedona, Lars Hoffmann, Reinhold Spang, Gabriele Cavallaro, Sabine Griessbach, Michael Höpfner, Matthias Book, and Morris Riedel

Download

Final revised paper (published on 08 Jul 2020)
Preprint (discussion started on 30 Jan 2020)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Review on ACP Manuscript No. ACP-2019-481', Anonymous Referee #2, 19 Feb 2020
- AC1: 'Response to reviewer', Rocco Sedona, 23 Apr 2020
RC2: 'Referee comment on amt-2019-481', Anonymous Referee #1, 26 Feb 2020
- AC2: 'Response to reviewer', Rocco Sedona, 23 Apr 2020

Peer-review completion

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

AR by Rocco Sedona on behalf of the Authors (23 Apr 2020) Manuscript

ED: Referee Nomination & Report Request started (27 Apr 2020) by Christian von Savigny

RR by Anonymous Referee #2 (05 May 2020)

RR by Anonymous Referee #1 (05 May 2020)

Suggestions for revision or reasons for rejection

The authors have clearly adressed the revisions and improved the manuscript. The following points should be clarified prior to publication:

1.19 sentence should be revisited: denitrification and heterogenous activation both affect ozone loss but are different things
2.19 Voigt et al. (2018) use airborne remote sensing, which does not fall in the categories in situ or microwave
8.13ff More explanation is required for Fig 2. Should the reader focus on rows/columns, or the diagonal? or maximum values only? What about the "W" features? They are missing in the discussion and labels
8.14f 820-840 cm-1 region is discussed already in previous sentence; what about the very strong feature at BT index ca. 10?
9.2 R3 seems relatively important and R5 unimportant, both opposite to Fig. 2a. Should this be discussed? Could this explain the better performance of RF?
9.10 BT indices should be provided for the 790 and 1235 cm features to allow locating these features in Fig. 2
11.10 In the seasonal analysis, only the classes sts, ice and nat are shown. Does this mean that also mixtures containing sts are summarized as "sts" etc.? Is "nat_sts" counted as both "nat" and "sts", thus both in Fig 10a and 12a?
11.30 A recent study indicates that large aspherical "beta-NAT" particles can be discriminated in ir spectra (Woiwode et al., 2019). Are the methods
used here sensitive to such particles?
12.11ff Which part of the subsequent section belongs to Fig 15. and which to Fig. 16? Is Fig. 16 discussed at all? The discussion is short for 2 figures including 6 panels in total
12.13 "almost all" seems too strong
12.16 Do percentages in text refer to y-axes in Fig. 15? "nat_small" seems red >80%, while the text says "around 50 % and 60 %". Some explanation is required on how percentages given in the text are related to the figures

References:
Woiwode et al., Geophys Res. Lett., 46, 13,420–13,429, 2019

Technical
3.1 "((" and "))"
7.31 numbers of figures should be ordered according to appearance in text; "pipeline" Fig 4 = Fig 2?
15.6 2010â€“2011
Fig 2 suggest to move "R" labels to middle of respective windows; otherwise one may get the impression that R1 etc are lines only; increments of y and x-axis should be identical
Fig 4 2x"radianCES"; explain "f.e."
Fig 5 grey/dark grey shading should be labelled or defined in caption
Fig 15,16 define N

Hide

ED: Publish subject to minor revisions (review by editor) (12 May 2020) by Christian von Savigny

AR by Rocco Sedona on behalf of the Authors (20 May 2020) Author's response Manuscript

ED: Publish subject to technical corrections (15 Jun 2020) by Christian von Savigny

AR by Rocco Sedona on behalf of the Authors (17 Jun 2020) Manuscript

Short summary

Polar stratospheric clouds (PSCs) play a key role in polar ozone depletion in the stratosphere. In this paper, we explore the potential of applying machine learning (ML) methods to classify PSC observations of infrared spectra to classify PSC types. ML methods have proved to reach results in line with those obtained using well-established approaches. Among the considered ML methods, random forest (RF) seems to be the most promising one, being able to produce explainable classification results.