the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 31 Jul 2026
Investigating the cloud liquid water and droplet concentration relationship across cloud morphologies with machine learning
Abstract. The relationship between cloud droplet number concentration (Nd) and cloud liquid water path (๐) is investigated in the context of detailed information about cloud morphology derived from self-supervised machine learning (ML) interpretation of satellite imagery over the North Atlantic ocean. While different cloud morphologies occupy distinct regions in the ๐-Nd phase space, the “inverted-V” relationship observed between ๐ and Nd does not arise as a result of their distribution in this space, and most cloud morphologies independently exhibit the inverted-V pattern to some degree. A novel approach to investigate this relationship is demonstrated using continuous vector space image embedding representations of cloud morphology produced by the self-supervised ML. Analysis of non-linear regression between these cloud morphology representations, Nd, and ๐ indicates that while cloud morphology information can explain most of the variance in ๐ between cloud scenes, the portion of ๐ variance that can be explained by Nd is mostly independent of cloud morphology and explains the inverted-V pattern.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Chemistry and Physics.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(9210 KB) - Metadata XML
-
Supplement
(23597 KB) - BibTeX
- EndNote
Status: open (extended)
- RC1: 'Comment on egusphere-2026-4426', Michael Diamond, 08 Sep 2026 reply
Data sets
Code and Data Supplement for: "Investigating the cloud liquid water and droplet concentration relationship across cloud morphologies with machine learning" Andrew Geiss https://zenodo.org/records/18237664
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 154 | 77 | 18 | 249 | 40 | 18 | 20 |
- HTML: 154
- PDF: 77
- XML: 18
- Total: 249
- Supplement: 40
- BibTeX: 18
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In this paper, the authors apply an unsupervised ML classification algorithm (albeit manually consolidated ex post) to test whether the classic โinverted-Vโ pattern in Nd-LWP space, which is suggestive of regime-dependent secondary indirect effects, can be explained by the convolution of different cloud morphologies. It cannot, which they show convincingly with evidence that the inverted-V recurs within each morphology regime and that the neural network can explain most LWP variance with morphology information alone but needs Nd to recreate the inverted-V pattern. The manuscript is very well written, the figures are clear, and the inclusion of a continuous representation of morphology and the differentiation analysis on the neural network are novel and interesting. Iโm tempted to recommend publication as-is, but do have some minor comments below the authors may wish to consider before publication. -Michael Diamond
General comment:
Assumed direction of causality: The analysis is premised on wanting to use Nd to predict LWP, but how do you account for, e.g., the possible influence of LWP on Nd (ala pathway E3 in Gryspeerdt et al., 2019)? I also struggle with how to express the right amount of uncertainty over causality in discussing the Nd-LWP relationship and inverted-V without making a text completely unreadable, but the authors could address this more head-on in the introduction.
Specific comments: