Interpretable Machine Learning Quantifies Composition and Size Controls on Aerosol Spectral Absorption

Wang, Wenfang; Tian, Pengfei; Zeng, Shuhua; Zhang, Yifei; Yu, Zeren; Cui, Chen; Wu, Yunfei; Chen, Min; Zhang, Lei

doi:10.5194/egusphere-2025-6118

Preprints

https://doi.org/10.5194/egusphere-2025-6118

Preprints

09 Jan 2026

| 09 Jan 2026

Status: this preprint is open for discussion and under review for Atmospheric Chemistry and Physics (ACP).

Interpretable Machine Learning Quantifies Composition and Size Controls on Aerosol Spectral Absorption

Wenfang Wang, Pengfei Tian, Shuhua Zeng, Yifei Zhang, Zeren Yu, Chen Cui, Yunfei Wu, Min Chen, and Lei Zhang

Abstract. The spectral dependence of aerosol absorption, characterized by the absorption Ångström exponent (AAE), strongly influences radiative effects, yet the relative importance of controlling factors remains poorly quantified. We integrate multisource observations with an interpretable machine-learning framework (Shapley Additive Explanations, SHAP) to disentangle the roles of chemical composition and particle size in shaping AAE and to evaluate radiative impacts. Field observation in Beijing reveal that near-surface AAE is predominantly influenced by higher fine mineral dust and water-soluble inorganic ions fractions. Multi-year columnar data identify dust loading as the dominant factor, followed by carbonaceous aerosols. The fine-mode radius accounts for 29 % of size parameters cumulative importance and ranks closely with black carbon. SHAP diagnostics highlight that columnar AAE contributes to radiative forcing at the top of the atmosphere (TOA) comparably to single scattering albedo (SSA), while its impact is clearly weaker at the bottom of the atmosphere and in the atmosphere. These findings help clarify AAE determinants and reduce uncertainties in aerosol radiative effect assessments.

Received: 08 Dec 2025 – Discussion started: 09 Jan 2026

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.

Download & links

Preprint (PDF, 1614 KB)

Supplement (1763 KB)

Download & links

Wenfang Wang, Pengfei Tian, Shuhua Zeng, Yifei Zhang, Zeren Yu, Chen Cui, Yunfei Wu, Min Chen, and Lei Zhang

Status: open (until 27 Feb 2026)

Post a comment Subscribe to comment alert

CC1:
'Comment on egusphere-2025-6118', Xiyao Chen, 21 Jan 2026 reply
This paper conducts a detailed study on the influencing factors of ground and atmospheric column AAE, revealing distinct patterns of influence. However, it is evident that there is a lack of correlation and comparison between the studies on the ground and atmospheric column. Furthermore, a significant portion of the text is devoted to discussing the SHAP of TOA, ATM, and BOA with respect to atmospheric column optical properties. However, are TOA, ATM, and BOA calculated based on atmospheric column optical properties and radiative transfer models? Therefore, SHAP is likely just a data-driven decomposition and description of the traditional radiative transfer simulation process. In summary, I suggest a major revision.

Major comments:
It is necessary to enhance the comparison between ground and atmospheric column AAE, especially to explain why carbonaceous aerosols and dust are not the main driving factors on the ground.

The article initially employs many correlations, followed by regression analysis using Multiple Linear Regression (MLR) for ground-based AAE, while the atmospheric column is studied using seven interpretable machine learning models for AAE and six radiative factors. There is considerable overlap in data analysis functions among correlations, MLR, and machine learning. Firstly, why isn't interpretable machine learning used for ground-based AAE? The same method makes it easier for readers to compare driving patterns between ground and column AAE.

Why does the ground-based MLR directly state in the methods section that black carbon, brown carbon, and dust are not included or as the “intercept term” (Lines 179-180)? The apparent reason for exclusion may stem from the correlation analysis in the results section (e.g., Section 3.2). I am not denying your viewpoint, on the contrary, I find it very interesting. Unfortunately, the author did not focus on analyzing possible mechanisms and instead conducted extensive repetitive analysis and modeling.

The interpretable machine learning for the six radiative quantities seems (e.g., Figs. 6-7) to merely describe the traditional radiative transfer simulation process via a data-driven machine learning method. It is suggested to move it to the Supplementary Information (SI), with only the main conclusions appearing in the main text. Is this comment correct? Perhaps you need to clarify how radiation variables are obtained, rather than just relying on literature, why machine learning is applicable, and what new discoveries machine learning has made.

Minor comments:
Lines 44-46: Provide AAE for dust and brown carbon separately.

In the Introduction, it is necessary to separately review the ground and column AAE, with fewer or unclear reviews of column AAE. Why is it necessary to study both ground and column AAE simultaneously?

Line 180, 304-307, 323-325, and so on: These are key points, but too simple. Need more discussions.

Lines 251-254, Fig. 1: It is not clear or direct for reads to understand! Perhaps Figure 1 contains a lot of information, but lacks visualization skills, making it difficult to highlight key points.

MLR and machine learning first need to report R2 and RMSE, such as in Lines 313-314. These are the most direct indicators to win the trust of readers. Please check the whole paper.

Lines 336-338: Organize sections around this point.

Reply
Citation: https://doi.org/10.5194/egusphere-2025-6118-CC1
RC1:
'Comment on egusphere-2025-6118', Anonymous Referee #1, 31 Jan 2026 reply
Review of "Interpretable Machine Learning Quantifies Composition and Size Controls on Aerosol Spectral Absorption"
This manuscript by Wang et al. investigates the complex drivers of the Absorption Ångström Exponent (AAE) using a combination of high-resolution field observations in Beijing and multi-year AERONET columnar data. The study addresses a persistent challenge in atmospheric science: quantitatively disentangling the relative importance of chemical composition versus particle size in determining aerosol spectral absorption. The authors employ Shapley Additive exPlanations (SHAP) to provide a ranked attribution of AAE drivers. The study quantifies how AAE variations directly influence aerosol radiative forcing efficiency (ARFE), finding that AAE is a primary driver of cooling efficiency at the Top of the Atmosphere (TOA). The manuscript is well-structured and the methodology is fairly robust, with the caveats detailed below.
Major issues
The study relies heavily on SHAP to attribute drivers of AAE which identifies associations, not necessarily causal links. The authors acknowledge nonlinearity and collinearity among predictors, however, if two variables are highly correlated (e.g., $D_{APS}$ and FMD mass fraction, $r=0.64$), SHAP can sometimes "split" the importance between them in ways that don't reflect physical reality. The manuscript would benefit from a more explicit discussion on whether the "importance" found by the model is supported by Mie theory or other physical optical models to bridge the gap between statistical importance and physical causation.

Near-surface chemical data was collected via offline filter sampling (day/night blocks), while optical data was online (hourly). While the authors matched these temporally, the coarse resolution of the chemical data (12-hour averages) likely masks the fine-scale diurnal variability seen in the optical $AAE_{sfc}$.

The authors describe using a ‘consistent split’ for model training, testing and validation but never explicitly state what that is. Please describe the method for splitting and why that’s appropriate for this dataset.

Minor issues
The authors focus on a specific site in Beijing, which given the observational constraints seems reasonable, but it would be beneficial to include a sentence or two in the conclusion explicitly stating how these results might change (and therefore their relevance) in cleaner or more dust-dominant global regions.

Please mention the ranges quoted in the results section. Are these 1 standard deviation, or something else?

L217, "coare-mode" should be corrected to "coarse-mode".

Reply
Citation: https://doi.org/10.5194/egusphere-2025-6118-RC1

Wenfang Wang, Pengfei Tian, Shuhua Zeng, Yifei Zhang, Zeren Yu, Chen Cui, Yunfei Wu, Min Chen, and Lei Zhang

Supplement

https://doi.org/10.5194/egusphere-2025-6118-supplement

Wenfang Wang, Pengfei Tian, Shuhua Zeng, Yifei Zhang, Zeren Yu, Chen Cui, Yunfei Wu, Min Chen, and Lei Zhang

Viewed

Total article views: 239 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
156	69	14	239	34	9	12

HTML: 156
PDF: 69
XML: 14
Total: 239
Supplement: 34
BibTeX: 9
EndNote: 12

Views and downloads (calculated since 09 Jan 2026)

Month	HTML	PDF	XML	Total
Jan 2026	156	69	14	239

Cumulative views and downloads (calculated since 09 Jan 2026)

Month	HTML	PDF	XML	Total
Jan 2026	156	69	14	239

Viewed (geographical distribution)

Total article views: 213 (including HTML, PDF, and XML) Thereof 213 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 31 Jan 2026

Short summary

We separated the roles of chemical composition and particle size in shaping absorption Ångström exponent (AAE) using ground and column measurements together with an interpretable machine learning. We found that near surface AAE is governed by higher fine mineral dust and inorganic ions fractions. Fine-mode effective radius has an influence close to black carbon on columnar AAE. Columnar AAE contributes to radiative forcing at the top of the atmosphere comparably to single scattering albedo.


Total:	0
HTML:	0
PDF:	0
XML:	0