Understanding the drivers of marine liquid-water cloud occurrence and properties with global observations using neural networks

Andersen, Hendrik; Cermak, Jan; Fuchs, Julia; Knutti, Reto; Lohmann, Ulrike

doi:https://doi.org/10.5194/acp-17-9535-2017

Articles | Volume 17, issue 15

https://doi.org/10.5194/acp-17-9535-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/acp-17-9535-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 17, issue 15

Research article

|

08 Aug 2017

Research article |

| 08 Aug 2017

Understanding the drivers of marine liquid-water cloud occurrence and properties with global observations using neural networks

Hendrik Andersen, Jan Cermak, Julia Fuchs, Reto Knutti, and Ulrike Lohmann

Download

Final revised paper (published on 08 Aug 2017)
Preprint (discussion started on 31 Mar 2017)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Review: Andersen 2017 ACPD', Anonymous Referee #1, 12 Apr 2017
- AC1: 'Responses to Referee 1', Hendrik Andersen, 23 May 2017
RC2: 'Review of 'Understanding the drivers of marine liquid-water cloud occurrence and properties with global observations using neural networks' by Andersen et al.', Anonymous Referee #2, 04 May 2017
- AC2: 'Responses to Referee 2', Hendrik Andersen, 23 May 2017
RC3: 'Review of "Understanding the drivers of marine liquid-water cloud occurrence and properties using neural networks.', Anonymous Referee #3, 26 May 2017
- AC3: 'Responses to Referee 3', Hendrik Andersen, 12 Jun 2017

Peer-review completion

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

AR by Hendrik Andersen on behalf of the Authors (12 Jun 2017) Author's response Manuscript

ED: Referee Nomination & Report Request started (12 Jun 2017) by Barbara Ervens

RR by Anonymous Referee #1 (15 Jun 2017)

RR by Anonymous Referee #2 (19 Jun 2017)

Suggestions for revision or reasons for rejection

I thank the authors for their work in responding to the comments. The majority of my comments have been addressed, only two things remain. I would suggest the authors include these at their own discretion. However, some mention of them might improve the paper still further.

The authors have been clear that they would prefer not to include results from a single global ANN. While creating a single relationship for predicting oceanic CLF globally may be difficult, this is not as far-fetched as they make it sound, as this is the central aim of a cloud parametrisation. By making their ANN regionally dependent, this is similar to including the latitude and longitude in a parametrisation. I understand that it is difficult to include all of the necessary parameters in a single ANN to predict the CLF (or other parameters), but as with a cloud parametrisation, the regions or conditions where the global ANN is deficient would indicate locations for future research.

Secondly, I think that my previous point about overlying ice cloud was misunderstood. The CLF is the fraction of retrievals where clouds with liquid tops are detected, so any situation where there is overlying ice cloud has the potential to reduce the CLF without changing the properties of the underlying liquid cloud. If there are not suitable variables in the ANN to predict the ice cloud fraction, it essentially becomes random noise and would thus limit the predictive ability of the ANN. Using only single-layer cloud retrievals does not address this issue, as they provide only the fraction of retrievals where a single layer liquid water cloud is detected, a subset of CLF. The only way I can see to easily address this issue is to restrict the study to gridboxes where only liquid cloud is detected, ensuring that there is no overlying ice cloud to artificially reduce the CLF. While excluding pixels with ice clouds is not essential (as the ANN clearly already has significant predictive ability), I would recommend that the authors consider it as a method of improving the statistics generated by the ANN.

Hide

ED: Reconsider after minor revisions (Editor review) (27 Jun 2017) by Barbara Ervens

AR by Hendrik Andersen on behalf of the Authors (06 Jul 2017) Author's response Manuscript

ED: Publish as is (10 Jul 2017) by Barbara Ervens

AR by Hendrik Andersen on behalf of the Authors (11 Jul 2017)

Short summary

Aerosol-cloud interactions continue to contribute large uncertainties to our climate system understanding. In this study, we use near-global satellite and reanalysis data sets to predict marine liquid-water clouds by means of artificial neural networks. We show that on the system scale, lower-tropospheric stability and boundary layer height are the main determinants of liquid-water clouds. Aerosols show the expected impact on clouds but are less relevant than some meteorological factors.