A Long Short-Term Memory (LSTM) Network for Hourly Estimation of PM_2.5 Concentration in Two Cities of South Korea

Round 1

Reviewer 1 Report

The review of the manuscript titled: "A long short-term memory (LSTM) network for hourly estimation of PM_2.5 concentration in two cities of South Korea".

The paper contains well-decribed analysis of the use of six known machine learning models to predict hourly PM_2.5 concentration based on one-year data from 8 measuring stations in two cities. The manuscript is properly constructed, the methods are well described, the model creation process is clearly described.

However, I have a few doubts and comments.

1. On what set were the match quality indicators described in section 3.4 determined? All or just a test?

2. The test set was not randomly selected only picked a'priori as the last part of the year. What impact can a particular selection of a test set have on the quality assessment of matched models? The authors themselves write that "high fine dust periods that usually occurs in spring and winter seasons". How do these seasons relate to the research area (in what months are they?) And how is the climate there. Please remember that readers are from different continents.

3. How have the best parameters, described in section 3.3.1, been determined? Was it a method of minimizing the take-off function? If so, which one? Or have you created several models with the given values based on experience and selected the model that best fits your data? This should be explained in the manuscript.

4. The literature review part it is worth supplementing with more current items on the use of machine learning for modeling of PM_2.5 concentrations, for example:

Kamińska J.A., The use of random forests in modelling short-term air pollution effects based on traffic and meteorological conditions: a case study in Wrocław, Journal of Environmental Management 217C (2018) pp. 164-174

Li, Xie, ren et al. Urban PM2.5 concentration prediction via attention-based CNN-LSTM, Appl. Sci. 2020, 10, 1953

5. Bearing in mind the wide spectrum of readers, each abbreviation appearing in the text should also be explained, also AI (line 28) and ML (line 40).

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

As particulate matter is a dangerous kind of air pollution, the research about it displays some interest. Especially when the World Health Organization and the International Agency of Research Center categorize it as carcinogenic and a major cause of allergies, pulmonary and cardiovascular diseases, morbidity and mortality. Nevertheless, the model implemented in the research is purely and utterly deterministic. That draws a twofold observation to be considered or added into the study. First, the evaluation and the differences between the predicted values and the real ones. Based on it, the changes needed in the assumptions for a stochastic model.   

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thank you for taking this review seriously. The explanations satisfy me. I believe that the manuscript additions are sufficient.
About the unfortunate times of the year. I meant in December it is cold in one part of the world, warm in another, and not all readers are fluent in this knowledge. One of the main sources of dust is domestic heating (cold season) or forest fires (hot season). For the future, please comment briefly on this issue, because it is of great importance due to the seasonality of this phenomenon. It is obvious that Seoul is the capital of South Korea.

I accept manuscrict in its current version.