the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Aug 2025
Machine Learning Reveals Strong Grid-Scale Dependence in the Satellite Nd–LWP Relationship
Abstract. The relationship between cloud droplet number concentration (Nd) and liquid water path (LWP) is highly uncertain yet crucial for determining the impact of aerosol-cloud interactions (ACI) on Earth's radiation budget. The Nd-LWP relationship is examined using a machine learning (ML) random forest model applied to five years of satellite data at grid resolutions ranging from 10° to 0.05° in 12 distinct regions. In the subtropics, the shape of the Nd-LWP relationship switches from an inverted-V at 1° grid-resolution to an "M" shape at 0.1° resolution with decreased dlnLWPdlnNd sensitivity. Tropical and midlatitude regions generally show a more positive sensitivity. Cloud sampling and filtering also influence this slope, wherein the exclusion of thin clouds, as commonly performed to reduce retrieval uncertainty, leads to strongly negative sensitivity across all regions. Precipitation is primarily responsible for driving the strength of the sensitivity, with strong positive slopes in raining clouds and negative and/or neutral responses found in non-raining clouds. A new method to compute radiative forcing from the ML model shows a robust Twomey radiative forcing across all regions and grid resolutions. However, LWP and cloud fraction rapid adjustments, which are ~50 % or smaller than the Twomey effect, decrease to negligible values with higher spatial resolution data. As Earth system models move toward higher spatial resolutions in the future, evaluating the LWP and CF adjustment contributions to the radiative forcing budget at these finer resolutions will be essential for evaluation and model development.
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
(7698 KB) - Metadata XML
-
Supplement
(11505 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-3850', Anonymous Referee #1, 24 Sep 2025
-
RC2: 'Comment on egusphere-2025-3850', Anonymous Referee #2, 25 Sep 2025
General comments:
In the article “Machine Learning Reveals Strong Grid-Scale Dependence in the Satellite Nd –LWP Relationship”, the authors employ a machine learning model to investigate the relationship between cloud droplet number concentration and liquid water path, and their connection to aerosol-cloud interactions, such as the Twomey effect.
With their random forest model, the authors provide highly interesting results for distinct changes of the Nd-LWP relationship with grid resolution and regional effects. As such, the article provides an efficient and innovative approach to quantify effects of aerosols on cloud processes, offering exciting opportunities for Earth system models.
Overall, the authors present their findings clearly and concisely, allowing readers to easily follow their approach. Hence, I regard this article with its findings on aerosol-cloud interactions and the introduced machine learning approach as a valuable contribution to the scientific community and future research. While I recommend this article for publication, I have some minor comments where additional clarification would be appreciated before publication.
Specific Comments:
L. 51-52: “We have generated a series of collocated global datasets at a series of spatial resolutions from 10° × 10° down to 0.05° × 0.05°”. Could you specify your resolutions in this section? The information can be found in Section 3.1, but it would be helpful to include a list of resolutions here.
L. 74-77: The naming of the filters (Q06, G18) does not seem intuitive to me. What do Q06 and G18 stand for? Please add either a reference or introduce the acronyms.
L. 126: “following 13 predictor variables”: Instead of only naming all variables, you could help the reader by providing an overview table for included predictors and their respective sources.
L. 131-132: “Each tree is trained on approximately 60% of the training dataset with replacement, utilizing the remaining 40% as out-of-bag observations to test tree performance”. Please describe how you split the dataset (random, temporal, spatial). Did you use the same dataset for validation and test? Ideally, you would have three datasets to ensure evaluating on an independent test set the model has not seen before.
L. 133-134:” evaluated using different hyperparameter values, such as the number of trees and the minimum number of samples per leaf”. It would be great to have an overview table for all hyperparameters.
L. 217: I found it a bit difficult to follow the results in this section. They mostly relate to different regional characteristics (i.e., California), but I miss a more general evaluation of the Nd-LWP relationship and a comparison across regions. Do you have the same number of samples in all regions? If not, it would be interesting to compare the findings between regions in connection to their robustness.
Figure 1: Add ENA and WNP in the figure caption.
Figure 7: Is this also for California, or averaged over all regions?
Citation: https://doi.org/10.5194/egusphere-2025-3850-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,786 | 50 | 11 | 1,847 | 15 | 22 | 23 |
- HTML: 1,786
- PDF: 50
- XML: 11
- Total: 1,847
- Supplement: 15
- BibTeX: 22
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments:
In this study, the authors investigate the Nd–LWP relationship (Nd: cloud droplet number concentration; LWP: liquid water path) retrieved from satellite observations at grid resolutions ranging from 10° to 0.05°. To reduce retrieval errors, they introduce a machine learning (ML) random forest model to estimate LWP using relevant cloud-controlling factors. After obtaining reliable ML results, the authors re-examine the Nd–LWP relationship and identify the main controlling factors that determine its characteristic shapes. They further apply this method to evaluate radiative forcing.
The reviewer is impressed by the methodology developed in this work, particularly the application of ML techniques to decompose the dominant controlling factors shaping the Nd–LWP relationship. The authors test their approach comprehensively across multiple grid resolutions (10° to 0.05°) and 12 different oceanic regions, successfully identifying general characteristics of the Nd–LWP relationship and its impact on radiative forcing. The reviewer finds this study innovative and believes it highlights a promising research direction for analyzing high-resolution satellite data. Therefore, the reviewer recommends publication of the paper, subject to minor revisions.
Many supplemental figures are shown in a separate file. In principle, the text should be readable without referring to the supplemental material. In this sense, it is better to place Figures S2 and S10 in the main text. Please reconsider the choice of the figures in the main text and those of the supplemental material.
Specific comments:
L75–76: Please explain the names of the filters, Q06 and G18. Do they refer to specific papers?
L140–141: CEP is located in the Eastern Pacific. The naming of this oceanic region could be improved. Why was no region in the Western Pacific selected? For example, the Western Pacific near the equator at around 160°E. It is a typical convective area.
L149, “Precipitation rates are also large in the tropics (compared to the subtropics).” According to Fig. 1e, the precipitation rate is larger in ST compared to TR.
L237, “precipitation can also decrease LWP”: What type of case is considered when precipitation leads to a decrease in LWP?
L304, Section 5.3: According to Fig. 6, relative humidity above PBL is not an important factor determining LWP. The discussion in Section 5.3 seems redundant.
Section 5.4 is understandable, but it can be improved in terms of readability. Please relate each cloud and radiative effect listed in Table 2 to the mathematical expression in the text.
L454: What does “the rapid adjustments” refer to here?