A Preliminary Fuzzy Inference System for Predicting Atmospheric Ozone in an Intermountain Basin

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study is based on the use of Fuzzy logic. This method has been implemented on several occasions to predict air pollution, including indoor pollution. This work illustrates in great detail the technique used by the referee to express doubt about the choice of physical parameters in the model.

First, for those who are not familiar with the Uinta Basin, it is appropriate to briefly describe its main characteristics related to air pollution. Furthermore, in figure 1, it would be appropriate to indicate the size of the scale or, better, to indicate the size of the circle drawn there.

The aspects to be reviewed in this work are essentially two. The first refers to the fact that the technical details presented in the paper bring the work to a very high length and therefore to the need for its reduction in size without altering the meaning of the research presented.

The most important criticism, however, is that the site under study (Uinta Basin) has very few high-ozone events. In fact, in figure 12 it is clearly seen that the only ozone episode that reached the quality standard was one related to the episode between February and March 2022. This single episode appears to be unrepresentative to a forecasting model.

The fact that once (2 Jan 2022) the model predicted high ozone without matching the real data clearly shows that some important parameters were not included in the model.

The four considered terms (snow cover, pressure, wind and solar irradiation) may not be sufficient to describe the phenomenon. For example, it seems appropriate to include actinic irradiation in the model rather than total irradiation. In addition, one of the most significant parameters could be the ground temperature or, even better, the temperature gradient at two different heights that would better describe the height of the mixed layer, as also reported in ref 2.

Regarding the failure to detect high ozone on January 2, it is worth mentioning that stakeholders are unlikely to accept a failure to predict high ozone, whereas they may be more inclined to accept a failure to not recognize an episode of high ozone.

In order to simplify the work, in figures 3 and 4, only the insets should be retained; however, in all figures, it would be appropriate to increase the font size to improve readability.

In conclusion, this work merits publication because it provides a basis for data processing that can be reproduced in other situations.

However, publication should be subject to a review in the terms indicated above.

Author Response

See attached PDF (identical for both reviewers)

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript presents a promising approach to predicting winter ozone levels in  Uinta Basin using fuzzy logic. To enhance its impact, the manuscript could benefit from a more detailed comparison with existing methods, clearer justification for some methodological choices, and a more rigorous evaluation of model performance. Strengthening the discussion on limitations and potential future improvements would also be valuable.

General comment

1.        simplify some jargon for a broader audience, or alternatively, provide a glossary of terms.

2.        The manuscript is generally well-organized, but there are sections where the flow could be improved, particularly in transitioning between different parts of the methodology.

Specific comment,

1.        Introduciton, it could benefit from a more detailed comparison with existing methods, particularly highlighting what makes the FIS approach novel or superior.

2.        In Methodology, about Data and Pre-processing, please clarify on how the representative values were derived, especially in terms of statistical handling. Also, the choice of variables like snow depth, wind speed, pressure, and solar insolation seems appropriate. However, more justification or comparison with alternative variables could strengthen this section.

3.        The use of synthetic examples to demonstrate the FIS's behavior is good. But,  including more discussion on the limitations of these examples, particularly in terms of their representativeness of real-world scenarios.

4.        The hindcast results from the winter of 2021/2022 are good, But discussing the potential impact of data sparsity on these results more explicitly and explore how this issue might be mitigated in future work.

5.        The evaluation metrics used to assess the FIS performance are not fully detailed. Adding standard metrics (e.g., accuracy, precision, recall) for model evaluation would help quantify the model’s performance more rigorously.

6.        The discussion is thorough, particularly in addressing the strengths and limitations of the FIS. However, the manuscript could benefit from a deeper exploration of why the model performed poorly in certain scenarios (e.g., the 2 January 2022 case) and what specific improvements could be made.

7.        While the stuidy mentions traditional NWP models, a more detailed comparison with these and other machine learning approaches would be beneficial. This could include a discussion of computational efficiency, ease of use, and explainability.

8.        Please, expand on the specific challenges anticipated in deploying the system and how these might be overcome.

Author Response

See attached PDF, identical for both reviewers

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The new version is better than the previous one. 

Reviewer 2 Report

Comments and Suggestions for Authors

Thanks for revising the manuscript with my comments. The manuscript has significantly improved and is suitable for publication.