NO2 Concentration Estimation at Urban Ground Level by Integrating Sentinel 5P Data and ERA5 Using Machine Learning: The Milan (Italy) Case Study
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper describes an application of Machine Learning for modelling/reproducing the NO2 concentration at ground level using TROPOMI NO2 tropospheric column data. This subject was very appealing a few years ago when TROPOMI was launched. NO2 product is one of the most popular, primarily when the study area is located in Europe.
The general remarks concerning methodology:
The authors did not comment on atmospheric mixing/stability and boundary layer height. Many papers presented that the tropospheric column is only sometimes directly related to the concentration at the surface. Please discuss this source of uncertainty in the paper. Also, consider using additional parameters from ERA5 (or derived from ERA5), which describe atmospheric vertical mixing as another supporting variable.
- May also the averaging kernel be included as an additional variable?
The authors did not comment on the qa_value threshold they assumed in their study. Was there any risk of including artefacts in machine learning?
- Table 3 – Testing data have a different mean than training data. Is there any particular reason for this lack of randomness? If you have a small dataset, consider using k-fold cross-validation (also available in scikit-learn).
- I did not find hyperparameters of the models used in this study. Please include them in the appendix. Did you use any automatic method for hyperparameter tuning?
Comments on the paper structure:
- The goal of the paper should be explicitly stated at the end of the introduction section
- Conclusions section needs to be included at the end of the manuscript. The author should once again confirm that the goal has been achieved.
The introduction section is too long and too general; what is the added value of Figure 2 in the scope of this paper?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe article by Cedeno-Jimenez and Brovelli presents research on atmospheric NO2 on the comparison between Sentinel-5P satellite data and in situ data from fixed stations. Furthermore, correlations and models are used.
I consider the article well written. English is fluid, I detected few errors and in some cases overly technical words.
I found errors in figures and tables, in some cases the captions did not coincide with what was shown in the figure or table. Furthermore, the author must include web pages and links where NO2 satellite and meteorological data can be downloaded in order for the article to be reproducible.
Line 26-30: two consecutive sentences begin with "on the other hand". Please rewrite.
Line 33: Is the author sure of the proposed limit of 200 ug/m3 per hour? I opened the reference and did not find the data reported in the text.
Line 45-46: please insert reference
Line 26-30: another case where the author uses "on the other hand" in two consecutive sentences. Please rewrite.
Line 73: what is ML?
Line 158: Sentinel 5P data… insert a link where these data are available. This work, like all published work, should be reproducible. Without this information the work cannot be reproduced. In every black dot the author must include link and/or web page.
Table 1: in the caption specify the difference between ρp and ρs.
Figure 4: in the figure are not presents the green marker. Please revise the caption.
Line 207-212: in my opinion this sentence is confusing. Why you need to download an image from WEKEO DIAS? Is it not possible to download the NO2 data together with the map on the TROPOMI main page? I don't understand why download a map.
Line 214: Why you replace data of ground sensors? The ARPA station don’t provide meteorological data? Moreover, provide a link of ERA5 data.
Line 249: (Figure 5).
Line 250: provide information of about the name of the chemiluminescence analyzers and model of this instrument.
Figure 7 and 8: low quality figure, the number in the axes are very small. Improve these figures.
Figure 9: What is the difference between the two figures? I think the caption is wrong.
Table 7: in this table RMSE and NRMSE (%) are not present. Please revise this table. Why you use Pearson correlation and R2? What additional information does R2 aggregate?
Conclusions are missing. Only it describes something in line 477-480. For this type of article and the topic covered a conclusion section must be included.
Comments on the Quality of English LanguageMinor editing english are detected.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for Authors1. Lines 101-114：They belong to the methods section, and it is recommended that the authors move to the “Materials and Methods” part.
2. The authors used satellite data and ground data in the manuscript, it is recommended to add websites that readers could access these data.
3. Lines 317-318： The authors divided the dataset into a training set (80%) and a testing set (20%) to validate the method proposed in this paper, and I suggest that the authors could us cross validation methods（such as 10-fold） to make the validation results more convincing.
4. Lines 332-339：Please list the formulas or methods corresponding to these models.
5. This manuscript seems incomplete, I suggest the authors add Conclusion part.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you for including my suggestions and comments.
The paper is now mature enough to be published.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe author improved the article. I consider the manuscript publishable.
Reviewer 3 Report
Comments and Suggestions for AuthorsAccept in present form
