the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Jun 2026
Classification and quantification of low-visibility events using deep learning over eastern China
Abstract. Low-visibility events (LVEs) threaten transportation safety and human health, yet accurately identifying them at a regional scale remains challenging. To address this, this study introduces a unified convolutional neural network (CNN) framework that integrates the geostationary satellite, meteorology, and fine particulate matter (PM2.5) observation data to identify all types of LVEs. The model can not only produce the spatiotemporal distribution of LVEs, including land and sea fog, but also quantify the intensity of LVEs linked to visibility reduction. By incorporating PM2.5, the polluted fog-haze over land can be discriminated from clean fog. Based on the model, we are able to investigate the environmental policy to reduce fog-haze and the corresponding population exposure. Over eastern China, a 20% reduction of overall PM2.5 can reduce the fog-haze area by 42% in winter. However, the corresponding population exposure is reduced less effectively by 15%, because the most populated region collocates with the most polluted region, where a further reduction of PM2.5 by over 40% is required to effectively reduce its population exposure. Here, the classified and quantified LVEs, along with the established relationship with pollution, explicitly guide the air-quality policy for the co-benefits of improving air visibility and human health.
Competing interests: Dantong Liu is an editor of ACP.
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
(6885 KB) - Metadata XML
-
Supplement
(3141 KB) - BibTeX
- EndNote
Status: closed
- RC1: 'Comment on egusphere-2026-2617', Anonymous Referee #2, 24 Jun 2026
- RC2: 'Comment on egusphere-2026-2617', Anonymous Referee #1, 29 Jul 2026
- AC1: 'Response to Referee #2 (RC1) and Referee #1 (RC2)', Dantong Liu, 26 Aug 2026
Status: closed
-
RC1: 'Comment on egusphere-2026-2617', Anonymous Referee #2, 24 Jun 2026
General Comments
This study proposes a unified convolutional neural network (CNN) framework that integrates geostationary meteorological satellite observations, meteorological reanalysis data, and PM2.5 observations to identify all types of low-visibility events. The framework shows good potential for quantifying the intensity and spatial extent of low-visibility events. In addition, this quantitative framework reveals the nonlinear responses of fog-haze events and population exposure to PM2.5 reduction, providing a useful reference for visibility improvement and air quality management. Overall, this study provides a valuable methodological contribution to the monitoring and assessment of low-visibility events. I suggest that the manuscript be accepted after the authors adequately address the specific comments below.
Specific Comments
- The discussion in the second paragraph of the Introduction on the development of fog identification methods needs to be further clarified, especially regarding the advantages of machine learning methods in nonlinear feature extraction and multi-source data fusion. In addition, I suggest further focusing on the necessity of integrating satellite spectral data, meteorological reanalysis data, and ground-based PM2.5 observations.
- In Sect. 2.4, the authors are suggested to explain how PM2.5 and visibility station data were interpolated or mapped onto the FY-4A grid, how missing values were handled, and whether the use of nearest-neighbor interpolation may introduce spatial uncertainty in regions with sparse monitoring stations.
- In Sect. 4.1, I suggest that the authors supplement the model performance for the classification of different types of low-visibility events.
- In the Results and Discussions section, several typical cases are selected for analysis. I suggest that the authors provide the basis for case selection and explain whether these cases are representative. In addition, if annual results are available, more complete seasonal statistics should be supplemented to enhance the robustness of the conclusions.
- In Fig. 4b, the authors compare the sea fog intensity simulated by the Sat model with the visibility reduction observed at Dinghai, Putuo, and Shengsi stations. However, the Sat model result in the figure seems to be shown as only one unified simulated value. Since the three stations are located at different spatial positions and the model output is gridded sea fog intensity, the simulated intensity should theoretically be extracted separately from the corresponding or nearest grid cells of each station for comparison.
- I suggest that the authors further discuss the reasons for the seasonal differences in the area of low-visibility events.
- In Lines 346-357, I suggest that the authors supplement quantitative results on the response of fog-haze area to PM2.5 reduction.
- In Lines 371-378, the authors show that reducing winter PM2.5 concentrations to 40% leads to simulated fog-haze areas that are highly consistent with the summer results. I suggest that the authors further clarify whether this sensitivity experiment controlled for or considered seasonal differences in meteorological conditions.
- In Line 404, the main text states that the “mean fog-haze area declined by 42%”, whereas the value labeled in Fig. 6e seems to be 43%. I suggest that the authors check the consistency between the value in the main text and that in the figure.
Technical Corrections
- Line 38: There is a formatting error with double parentheses in “transportation safety ((Ord Us Epa...”.
- The capitalization style of words in the section headings should be made consistent throughout the manuscript.
Citation: https://doi.org/10.5194/egusphere-2026-2617-RC1 - RC2: 'Comment on egusphere-2026-2617', Anonymous Referee #1, 29 Jul 2026
- AC1: 'Response to Referee #2 (RC1) and Referee #1 (RC2)', Dantong Liu, 26 Aug 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 195 | 108 | 44 | 347 | 60 | 27 | 35 |
- HTML: 195
- PDF: 108
- XML: 44
- Total: 347
- Supplement: 60
- BibTeX: 27
- EndNote: 35
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General Comments
This study proposes a unified convolutional neural network (CNN) framework that integrates geostationary meteorological satellite observations, meteorological reanalysis data, and PM2.5 observations to identify all types of low-visibility events. The framework shows good potential for quantifying the intensity and spatial extent of low-visibility events. In addition, this quantitative framework reveals the nonlinear responses of fog-haze events and population exposure to PM2.5 reduction, providing a useful reference for visibility improvement and air quality management. Overall, this study provides a valuable methodological contribution to the monitoring and assessment of low-visibility events. I suggest that the manuscript be accepted after the authors adequately address the specific comments below.
Specific Comments
Technical Corrections