Assessing the Effects of Urban Morphology Parameters on PM_2.5 Distribution in Northeast China Based on Gradient Boosted Regression Trees Method

Round 1

Reviewer 1 Report

COMMENTS

to manuscript sustainability-1578183 “Assessing the effects of urban morphology parameters on PM2.5 distribution in northeast China base on gradient boosted regression trees method“ by Peng Cui et al.

 

The authors investigated spatio-temporal distribution of PM2.5 over a limited area of 10 km², which includes urban blocks within the city of Harbin, in northeast China. Being located in a cold-climate zone, Harbin is characterized by complex environmental settings due to the following factors: 1) a huge consumption of fossil fuels for heating in winter; 2) agricultural biomass burning; 3) transportation pollution; and 4) long-range desert dust transport.

Investigation and prediction of spatio-temporal distribution of PM2.5 in such a polluted area is important for monitoring and future improvement of the environmental situation. This is essential for health of millions of people living in this highly polluted area. That is why I consider that the manuscript suits for publication in the Sustainability journal. However, the outcomes from this research may not be sufficiently representative. Some points require further clarification.

Below are my specific comments.

1. One of the main ideas of this study was to give advice on urban planning in order to improve environmental settings. In this respect the study appears to be of little scientific value, because the authors obtained their findings using a very limited set of measurements. Only a group of 21 days was selected from the six-month period of PM2.5 measurements from December 2020 to February 2020 and also from December 2021 to February 2021. Only measurements for 3 - 4 days were used for analyzing environment settings. Measurements for these rare days could represent only some random conditions. However, to give advice on urban planning, PM2.5 concentration should be analyzed for the full range of observed pollution conditions including extreme PM2.5 concentrations.
2. Specifically, from the selected 21 days of measurements, only 4 days were chosen to investigate the spatio-temporal distribution of daily average PM2.5 concentration. Why not use the whole set of 21 days to obtain a more reliable picture? The representativeness of the obtained spatio-temporal distribution of PM2.5 is of importance, as this is one of the main points of the study.
3. PM2.5 measurements for only 3 days, at one time moment (14:00), were used for analyzing correlation between PM2.5 and meteorological parameters. How representative was the obtained correlation in various meteorological conditions? The scarcity of measurements doesn’t meet general scientific standards for research papers. Why not use the whole set of 21 days at various hours which the authors have already collected? Please do it.
4. Please clarify why the correlation between PM2.5 and meteorological parameters was analyzed only at 14:00. It is essential to estimate this correlation at several time moments, when the maxima and minima of PM2.5 were observed, for example at 05:00, 10:00, 15:00, and at 22:00, in accordance with Fig.7.
5. According to Table 2, on the selected 3 days (20 December 2020, 9 January 2021, and 14 January 2021), PM2.5 concentration did not represent extreme pollution conditions. Specifically, Table 2 showed that PM2.5 ranged from 56 µg/m³ (5 Jan 2021, at 8 – 10 a.m.) to 263 µg/m3 (24 Jan 2021, at 8 – 10 a.m.). To be representative, the correlation between PM2.5 and meteorological parameters should be estimated for those extreme pollution conditions.
6. It is important to estimate correlation between PM2.5 and air temperature under various pollution conditions at each monitoring site separately for the whole set of 21 days with measurements.
7. I have a similar comment with respect to wind speed. At each monitoring site separately, what was the correlation between PM2.5 and wind speed for the whole set of 21 days under various polluted conditions? Please show.
8. For completeness, at each monitoring site separately, what was the correlation between PM2.5 and relative humidity for the whole set of 21 days? Please show.
9. Table 4: In the Table title, please clarify how these numbers in Table 4 were obtained, for what days and times.
10. Figure 10: Please clarify in the figure caption what real data were used in that comparison. To accurately evaluate a model capability of predicting spatio-temporal distribution of PM2.5, it is essential to compare model data with PM2.5 measurements at various locations and for various pollution conditions. Please do it as this is one of the main points of the study.

Author Response

Thank you for your detailed and constructive comments. Modifications have been made accordingly – all comments have been taken, as described below. (The modified article has been uploaded to the attachment. Please pay attention to the red part for specific modification.)

1. One of the main ideas of this study was to give advice on urban planning in order to improve environmental settings. In this respect the study appears to be of little scientific value, because the authors obtained their findings using a very limited set of measurements. Only a group of 21 days was selected from the six-month period of PM2.5 measurements from December 2020 to February 2020 and also from December 2021 to February 2021. Only measurements for 3 - 4 days were used for analyzing environment settings. Measurements for these rare days could represent only some random conditions. However, to give advice on urban planning, PM2.5 concentration should be analyzed for the full range of observed pollution conditions including extreme PM2.5 concentrations.

Thank you for your suggestion. The significance of this study is to explore the impact of urban morphology on the dispersion of air pollutants at block scale. We set up 25 points, covering most of the real building layout and urban morphology, to ensure that the test results are more accurate. There are few related literature. Most people start from urban scale or only study a few monitoring points, so our research perspective is different.

Affected by the test environment, Harbin has about 50 haze days a year. Among them, some snowfalls were uncontrollable, so it was excluded. In addition, some abnormal data were also excluded, so 21 days were finally selected.

In the study of temporal and spatial distribution, 3-4 days is considered as the method of sampling survey. In the temporal distribution test, 4 consecutive days were selected to consider the continuity of the test, while in the spatial distribution test, 3 days with different pollution levels were selected to cover different pollution conditions. However, your suggestion is more meaningful, so we reorganized and used the test data of 21 days.

 

2. Specifically, from the selected 21 days of measurements, only 4 days were chosen to investigate the spatio-temporal distribution of daily average PM2.5 concentration. Why not use the whole set of 21 days to obtain a more reliable picture? The representativeness of the obtained spatio-temporal distribution of PM2.5 is of importance, as this is one of the main points of the study.

Thank you for your suggestion. In the revised version, the days for testing was modified to 21 days.

 

3. 5 measurements for only 3 days, at one time moment (14:00), were used for analyzing correlation between PM2.5 and meteorological parameters. How representative was the obtained correlation in various meteorological conditions? The scarcity of measurements doesn’t meet general scientific standards for research papers. Why not use the whole set of 21 days at various hours which the authors have already collected? Please do it.

Thank you for your suggestion. In the revised version, the days for testing was modified to 21 days.

 

4. Please clarify why the correlation between PM2.5 and meteorological parameters was analyzed only at 14:00. It is essential to estimate this correlation at several time moments, when the maxima and minima of PM2.5 were observed, for example at 05:00, 10:00, 15:00, and at 22:00, in accordance with Fig.7.

We sincerely appreciate the valuable comments. In the revised version, the times for testing were modified to 5:00, 10:00, 16:00 and 22:00 respectively.

 

5. According to Table 2, on the selected 3 days (20 December 2020, 9 January 2021, and 14 January 2021), PM2.5 concentration did not represent extreme pollution conditions. Specifically, Table 2 showed that PM2.5 ranged from 56 µg/m3 (5 Jan 2021, at 8 – 10 a.m.) to 263 µg/m3 (24 Jan 2021, at 8 – 10 a.m.). To be representative, the correlation between PM2.5 and meteorological parameters should be estimated for those extreme pollution conditions.

Thank you for your suggestion. In the revised version, all of the days and the times for testing were modified.

 

6. It is important to estimate correlation between PM2.5 and air temperature under various pollution conditions at each monitoring site separately for the whole set of 21 days with measurements.

Thank you for your suggestion. In the revised version, the test times were modified to 5:00, 10:00, 16:00 and 22:00, and then their average values of 21 days were calculated for the final data analysis.

 

7. I have a similar comment with respect to wind speed. At each monitoring site separately, what was the correlation between PM2.5 and wind speed for the whole set of 21 days under various polluted conditions? Please show.

Thank you for your suggestion. In the revised version, the test times were modified to 5:00, 10:00, 16:00 and 22:00, and then their average values of 21 days were calculated for the final data analysis.

8. For completeness, at each monitoring site separately, what was the correlation between PM2.5 and relative humidity for the whole set of 21 days? Please show.

Thank you for your suggestion. In the revised version, the test times were modified to 5:00, 10:00, 16:00 and 22:00, and then their average values of 21 days were calculated for the final data analysis.

 

9. Table 4: In the Table title, please clarify how these numbers in Table 4 were obtained, for what days and times.

We express our sincere apologies for our vague expression and greatly appreciate the detailed comments you give. The urban morphology parameters of each measuring point under the buffer radius of 50m, 100m, 200m, 300m, 400m and 500m are extracted by GIS and divided into six groups for research. In each group, we successively analyzed the correlation between each urban morphology parameter and its corresponding PM_2.5 concentration at different times and different measuring points. Among them, the PM_2.5 concentration data comes from the 24-hour continuous monitoring of each measuring point for 21 days. In the revised version, this part has been supplemented.

 

10. Figure 10: Please clarify in the figure caption what real data were used in that comparison. To accurately evaluate a model capability of predicting spatio-temporal distribution of PM2.5, it is essential to compare model data with PM2.5 measurements at various locations and for various pollution conditions. Please do it as this is one of the main points of the study.

We express our sincere apologies for our wrong expression and greatly appreciate the detailed comments you give. The “ real value” should be “ actual value”. The data（actual value） in Figure 10 used for comparison with the predicted data were randomly selected from 30% of all measured data.

In this study, the measured PM_2.5 data of 21 days were taken from three periods of 8:00-10:00, 12:00-15:00 and 19:00-22:00 respectively, which is considered as the method of sampling survey. However, your suggestion is more meaningful, so we reorganized and used the 24-hour continuous measured data.

Reviewer 2 Report

This paper is an interesting topic. It is descriptive and analytical at the same time, providing interesting methodological ideas and specific results coherent with the research goals. Methodologically the paper the paper is clear and provides a useful way to deal with similar cases. However, this type of studies are no new.  There are existing studies that already assess the use of Machine Learning. You should justify this in the whole manuscript, including the conclusions. And also you need to include the existing papers that talk about this topic.

I have a general comment, I suggest to reducing a bit the content of the methodology. It’s a bit long. Also, I do not think the tittle of the manuscript is very aligned with the content of the study.  What you are doing here is to use Machine Learning to explore the nonlinear influence of urban morphology factors on PM2.5 concentration, and the tittle does not explain that.

 

In the Figures, you should add the source of the image.  For instance, Figure 4, does not seem the photo was taken in China. Also, in Figure 5, the scale you indicate on the top left is not the same to the scale of the circles. In all the graphs, indicate the parameters in the X and Y Axis. In many cases are missing and is difficult to read the content. Also I recommend to improving the quality of the images, in general are low quality.

Author Response

Thank you for your detailed and constructive comments. Modifications have been made accordingly – all comments have been taken, as described below. (The modified article has been uploaded to the attachment. Please pay attention to the red part for specific modification.)

1. This paper is an interesting topic. It is descriptive and analytical at the same time, providing interesting methodological ideas and specific results coherent with the research goals. Methodologically the paper the paper is clear and provides a useful way to deal with similar cases. However, this type of studies are no new. There are existing studies that already assess the use of Machine Learning. You should justify this in the whole manuscript, including the conclusions. And also you need to include the existing papers that talk about this topic.

Thank you for your suggestion. In the revised version, the application of machine learning in existing research has been added in Discussion, and their advantages and disadvantages were discussed. In addition, The cold-climate city in China have formed a unique situation of PM_2.5 dispersion due to their own climate conditions and urban layout. And the conclusions of existing studies are different. Therefore, it is necessary to fit the research results of local characteristics. This is also the significance of this study.

 

2. I have a general comment, I suggest to reducing a bit the content of the methodology. It’s a bit long. Also, I do not think the tittle of the manuscript is very aligned with the content of the study. What you are doing here is to use Machine Learning to explore the nonlinear influence of urban morphology factors on PM2.5 concentration, and the tittle does not explain that.

Thank you for your suggestion. In the revised version, the methodology was compacted and modified. 

 

3. In the Figures, you should add the source of the image. For instance, Figure 4, does not seem the photo was taken in China. Also, in Figure 5, the scale you indicate on the top left is not the same to the scale of the circles. In all the graphs, indicate the parameters in the X and Y Axis. In many cases are missing and is difficult to read the content. Also I recommend to improving the quality of the images, in general are low quality.

Thank you for your suggestion. In the revised version, all the figure problems were modified.

Reviewer 3 Report

This paper performs an investigation on the distribution of PM2.5 concentration in the city of Harbin to infer the impacts that different morphological areas of the city have on the flow field and pollutant concentration. The investigation is interesting but, in my opinion, lacks some solidity on the methods. Among other concerns expressed below, data from the deployed sensors should have been validated among “standard” instrumentations from the local Environmental Agency; this is fundamental especially when dealing with portable and non-standard instrumentations whose performance is oftentimes good in the laboratory but not as much in the open field. For this and the following reasons, I recommend major revisions.

 

Lines 80-81: if “prediction” is here used to address the forecast of pollutant concentration, modelling system as WRF with the Chemistry module (WRF-Chem) should be included in this list (see e.g., He et al 2022). If “prediction” also includes diagnostic studies of the distribution of pollutants in the urban area, other models can be included such as CFD (Jeanjean et al 2016), dispersion models (Gibson et al 2013) and hybrids/simplified fluid-dynamics models (Barbano et al 2020) which in different ways can include pollutant dispersion and morphological impacts on the flow and concentration.

Fig. 2: Where were the data collected? Is this a single-point monitoring station or an average of multiple monitoring stations within the city? In case of multiple sources, are they all similar (e.g., all referring to urban traffic conditions, such as having the station alongside major roads) or different (e.g., urban traffic and suburban background, with this last characterized by pollutant concentrations much more like the countryside rather than a city)? And how are they accounted for if so different?

Fig. 2: Are the concentrations shown as 24-hour averages? Please specify it.

Line 159 and y-axis label: use μ instead of u in the unit of measurement of concentrations. Please check for consistency throughout the paper.

Table 1: I would appreciate more information on the instrumentation deployed on the field. Regarding the weather station, what type of sensors do they include? It seems to be equipped with a wind turbine anemometer, but it is difficult to address from Fig. 4. What sensor is measuring temperature and humidity? Similarly, for the particle detector, I am wondering if it uses a mass or optic method to measure the PM2.5 concentrations.

Fig. 4: Due to their portability and I guess the ready-to-use nature of these instruments, can they be considered as low-cost sensors (e.g., Mao et al 2019, Brattich et al 2020)? In any case (but especially if they are low-cost sensors), a validation of these sensors among standard meteorological and air quality data (taken from the meteorological and air quality stations of the Environmental Agency) should be presented. Without this first step, I cannot be convinced that these sensors can be reliably used instead of standard meteorological and air quality sensors.

Table 3: Variables Li, A, and F within the third column are not defined, nor what the summary accounts for.

Sect 2.3.2: It is not clear to me what is the meaning of the buffer zones and how you determine their radii and centre. Please clarify.

Page 9: I think there are some malfunctioning of the page layout. Please check it.

Line 281: What is a loss function? What do you need it for?

Eqs 6,7,8: Variables within the equations are not defined.

Fig 7: It is quite difficult to discern each line; it may be worthy to compute the average and ensemble from these 25 measurement locations or separate the figure in more panels (e.g., in line with the description provided at lines 355-367).

Fig 8: The unit of measurement of PM concentration should be specified in the legend.

Fig. 7-8: If I am not mistaken, Figs 7 and 8 are evaluated from the 4-day average (hour by hour) at each monitoring point and plotted as a typical day evolution (Fig 7) and spatial distributions (Fig 8) from Fig. 7 concentrations at 10:00 and 22:00 respectively. If so, I cannot understand why the time series at 22:00 are always below 350 ug/m3 while the map legend has a 384 ug/m3 as a minimum. Please clarify.

Lines 377-381: What do you mean by meteorological temperature? Which is the difference with temperature? I have not fully understood this sentence: if you mean that changing WeaT provide a change in R2 without following the same increasing or decreasing path, I feel that using only 3 hours is not enough to support this conclusion. From Fig 9a I see a correlation between concentration and temperature with steeper trend slopes when the temperature is approaching -4°C and more gentle slopes below -10°C. Please clarify.

Lines 382-398: Same as for the previous comment.

Line 420 and Table 5: Should not MAE and MSE have a unit of measurement?

Fig. 11: What is the y axis representing? Should not be better to use a percentage of each factor?

Lines 484-497: Also more meteorological data should be included in this evaluation, since they are known to play a key role in pollutant dispersion, such as wind direction (determining the channelling of the flow in the urban texture), turbulence, heat and momentum fluxes (regulating the inflow-outflow of pollutant at the top of the street canyons), surface temperature (determining the thermal forcing of building and streets and regulating convection), solar radiation for shadowing effects and seasonality and others. Also, chemical factors should be included as PM2.5 is not only a primary but also e secondary pollutant. Please consider including some of these considerations in the discussion.

 

Reference

He, Z., Liu, P., Zhao, X., He, X., Liu, J., & Mu, Y. (2022). Responses of surface O3 and PM2. 5 trends to changes of anthropogenic emissions in summer over Beijing during 2014–2019: A study based on multiple linear regression and WRF-Chem. Science of The Total Environment, 807, 150792.

Jeanjean, A. P., Monks, P. S., & Leigh, R. J. (2016). Modelling the effectiveness of urban trees and grass on PM2. 5 reduction via dispersion and deposition at a city scale. Atmospheric Environment, 147, 1-10.

Gibson, M. D., Kundu, S., & Satish, M. (2013). Dispersion model evaluation of PM2. 5, NOx and SO2 from point and major line sources in Nova Scotia, Canada using AERMOD Gaussian plume air dispersion model. Atmospheric Pollution Research, 4(2), 157-167.

Barbano, F., Di Sabatino, S., Stoll, R., & Pardyjak, E. R. (2020). A numerical study of the impact of vegetation on mean and turbulence fields in a European-city neighbourhood. Building and Environment, 186, 107293.

Mao, F., Khamis, K., Krause, S., Clark, J., & Hannah, D. M. (2019). Low-cost environmental sensor networks: recent advances and future directions. Frontiers in Earth Science, 7, 221.

Author Response

Thank you for your detailed and constructive comments. Modifications have been made accordingly – all comments have been taken, as described below. (The modified article has been uploaded to the attachment. Please pay attention to the red part for specific modification.)

1. Lines 80-81: if “prediction” is here used to address the forecast of pollutant concentration, modelling system as WRF with the Chemistry module (WRF-Chem) should be included in this list (see e.g., He et al 2022). If “prediction” also includes diagnostic studies of the distribution of pollutants in the urban area, other models can be included such as CFD (Jeanjean et al 2016), dispersion models (Gibson et al 2013) and hybrids/simplified fluid-dynamics models (Barbano et al 2020) which in different ways can include pollutant dispersion and morphological impacts on the flow and concentration.

Thank you for your suggestion. WRF model is mainly used for the study of macro scale, but not for the micro scale of this study. However, your suggestion is more meaningful, so we have supplemented this part in the revised version. On the other hand, this study mainly focuses on the prediction of pollutant concentration. Therefore, we believe that the simulation research of CFD and others are not within the scope of research.

 

2. 2: Where were the data collected? Is this a single-point monitoring station or an average of multiple monitoring stations within the city? In case of multiple sources, are they all similar (e.g., all referring to urban traffic conditions, such as having the station alongside major roads) or different (e.g., urban traffic and suburban background, with this last characterized by pollutant concentrations much more like the countryside rather than a city)? And how are they accounted for if so different?

The data comes from the public data released by Heilongjiang meteorological observatory. It is comprehensively evaluated by the meteorological observatory according to the monitoring results of several meteorological stations. The weather stations are located in the suburb of Harbin. Because they are located in the suburbs, the monitoring values of several weather stations are basically the same. However, the measuring points of this study are set in the city, which are different due to the influence of buildings. This is also the focus of this paper.

 

3. 2: Are the concentrations shown as 24-hour averages? Please specify it.

The data comes from the daily PM_2.5 concentration released by Harbin meteorological observatory.

 

4. Line 159 and y-axis label: use μ instead of u in the unit of measurement of concentrations. Please check for consistency throughout the paper.

Thank you for your suggestion, in the revised version, all the writing errors were modified

 

5. Table 1: I would appreciate more information on the instrumentation deployed on the field. Regarding the weather station, what type of sensors do they include? It seems to be equipped with a wind turbine anemometer, but it is difficult to address from Fig. 4. What sensor is measuring temperature and humidity? Similarly, for the particle detector, I am wondering if it uses a mass or optic method to measure the PM2.5 concentrations.

Thank you for your suggestion. In the revised version, this part has been supplemented.

 

6. 4: Due to their portability and I guess the ready-to-use nature of these instruments, can they be considered as low-cost sensors (e.g., Mao et al 2019, Brattich et al 2020)? In any case (but especially if they are low-cost sensors), a validation of these sensors among standard meteorological and air quality data (taken from the meteorological and air quality stations of the Environmental Agency) should be presented. Without this first step, I cannot be convinced that these sensors can be reliably used instead of standard meteorological and air quality sensors.

Thank you for your suggestion. In the revised version, this part has been supplemented.

 

7. Table 3: Variables Li, A, and F within the third column are not defined, nor what the summary accounts for.

We express our sincere apologies for our oversight and greatly appreciate the detailed comments you give. In the revised version, the missing part was added.

 

8. Sect 2.3.2: It is not clear to me what is the meaning of the buffer zones and how you determine their radii and centre. Please clarify.

Thank you for your suggestion. In the actual neighborhood environment, diverse urban morphology factors have different degrees of impact on PM_2.5 concentration under different buffer zones. The significance of the buffer zone is to help obtain the maximum urban morphology factors that explain the variation of PM_2.5 concentration. The center of the circle in this study is 25 measuring points arranged in actual measurement, and the buffer radius is finally selected as 50-500m based on the actual situation.

This method is also used in existing references (Zhang, J., Cui, P., Song, H. (2020). Impact of urban morphology on outdoor air temperature and microclimate optimization strategy base on Pareto optimality in Northeast China, 180, 107035. ; Shi, Y., Xie, X., Fung, J. C. H., et al. (2018). Identifying critical building morphological design factors of street-level air pollution dispersion in high-density built environment using mobile monitoring. Building and Environment, 128, 248-259.).

In the revised version, this part has been supplemented.

 

9. Page 9: I think there are some malfunctioning of the page layout. Please check it.

Thank you for your suggestion. The format maybe changed during system submission process, in the revised version, it was improved.

 

10. Line 281: What is a loss function? What do you need it for?

Thank you for your suggestion. This study used the square loss function, which is very common in daily research. It is mainly used to solve the problem of regression analysis. The Gradient Boosted Regression Trees model obtains the final result by continuously minimizing the loss function. In the revised version, the missing part was added.

 

11. Eqs 6,7,8: Variables within the equations are not defined.

We express our sincere apologies for our oversight and greatly appreciate the detailed comments you give. In the revised version, the missing part was added.

 

12. Fig 7: It is quite difficult to discern each line; it may be worthy to compute the average and ensemble from these 25 measurement locations or separate the figure in more panels (e.g., in line with the description provided at lines 355-367).

Thank you for your suggestion. In the revised version, the presentation of Figure 7 has been changed.

 

13. Fig 8: The unit of measurement of PM concentration should be specified in the legend.

Thank you for your suggestion. In the revised version, it was corrected.

 

14. 7-8: If I am not mistaken, Figs 7 and 8 are evaluated from the 4-day average (hour by hour) at each monitoring point and plotted as a typical day evolution (Fig 7) and spatial distributions (Fig 8) from Fig. 7 concentrations at 10:00 and 22:00 respectively. If so, I cannot understand why the time series at 22:00 are always below 350 ug/m3 while the map legend has a 384 ug/m3 as a minimum. Please clarify.

We express our sincere apologies for our data mishandling and greatly appreciate the detailed comments you give. Meanwhile, In order to avoid the problem that the data were too large or too small, we changed the days of data used for analysis to 21 days of continuous monitoring data in the revised version.

 

15. Lines 377-381: What do you mean by meteorological temperature? Which is the difference with temperature? I have not fully understood this sentence: if you mean that changing WeaT provide a change in R2 without following the same increasing or decreasing path, I feel that using only 3 hours is not enough to support this conclusion. From Fig 9a I see a correlation between concentration and temperature with steeper trend slopes when the temperature is approaching -4°C and more gentle slopes below -10°C. Please clarify.

Thank you for your suggestion. “Meteorological temperature” is the temperature released by Harbin meteorological observatory, which represents the overall monitoring level of the whole region. “Temperature” is the temperature of each monitoring point in the urban block obtained through actual measurement. In the revised version, the days for testing was modified to 21 days. In addition, as for the slope problem you mentioned, we have recalculated and there is no obvious change, so we have not made further analysis.

 

16. Lines 382-398: Same as for the previous comment.

Thank you for your suggestion. In the revised version, this issue has been modified as comment 15.

 

17. Line 420 and Table 5: Should not MAE and MSE have a unit of measurement?

Thank you for your suggestion. In the revised version, all unit issues were modified.

 

18. 11: What is the y axis representing? Should not be better to use a percentage of each factor?

Thank you for your suggestion. In the revised version, it was corrected.

 

19. Lines 484-497: Also more meteorological data should be included in this evaluation, since they are known to play a key role in pollutant dispersion, such as wind direction (determining the channelling of the flow in the urban texture), turbulence, heat and momentum fluxes (regulating the inflow-outflow of pollutant at the top of the street canyons), surface temperature (determining the thermal forcing of building and streets and regulating convection), solar radiation for shadowing effects and seasonality and others. Also, chemical factors should be included as PM2.5 is not only a primary but also e secondary pollutant. Please consider including some of these considerations in the discussion.

Thank you for your suggestion. In the revised version, this part has been supplemented.

Round 2

Reviewer 1 Report

I consider that the authors properly addressed my comments. The revised manuscript is ready for publications.

Reviewer 2 Report

Thank you for the improvements.

Reviewer 3 Report

The authors have done a great job revising the article addressing all my comments and suggestions. I recommend this article to be accepted