This study was majorly inspired by Grange et al 2019 to use a random forest model (RF) to train, validate and predict the air quality concentration in a megacity of Argentina. Although the methodology is not new, this study has the potential to have a significant improvement to fill the data gap given the reason that some monitored trace gas concentrations were lacking but the pollutants are becoming a concern for local authorities. (line 145- 150) I would like to propose a few major revisions for the authors of this manuscript. After done with that improvement, I think it would be strong enough to publish on earth system science data:

1. Random forest models indeed are easy and quick to train. But if this study is only focusing on predicting the time series of pollution during COVID-19, there are better and more efficient machine learning models such as ARIMA compared with random forest. The key reason for many studies that authors mentioned in this article used RF is because it could provide key components reflecting the non-linear relationship among emissions, chemical reaction, and meteorological effects. Please see figure 3 in Grange et al 2019 and figure 1 in Yang et al. It is easy to generate that Gini importance either through python or R. Since this code the authors used were based on R, I would suggest they follow the code from Grange et al 2019 to generate the Gini importance plot. After getting those plots, you can compare your RF results with the reasons from previous literature to see if it makes sense. 

1. The authors also mentioned several times the meteorological impacts on air pollution. The earliest signal has been seen during COVID was the study in China from Le et al, 2020 which authors may consider mentioning this study result. Then it would also be beneficial to consider using RF models to generate new predictions by normalizing meteorological factors. This would give the third line in each panel of figure 3. You can do this by following the methodology in figure 5 of Grange et al 2019 and figure 2 of Yang et al 2021.  Vu et al 2019 made an additional improvement to weather normalization which you may also consider using this methodology. 

1. The description of details of why and how to interpret bivariate polar plots here is vague. The reason that Grange 2019 used this bivariate polar plot is that they wanted to prove that wind influence the dispersion of pollutants. Therefore it's better to show the meteorological impacts from the above suggestions that I mentioned first. Then applying a bivariate polar plot by combining with winds components if the meteorological impact is the dominant factor here. 

1. For the relationship between NO₂ and diesel please refer to Yang et al 2021. You can also consider normalizing other anthropogenic factors besides meteorology based on the result from the first suggestion. 

1. Please indicate how much data by # not percentage is used for training and how much data is used for validation/prediction. Due to some restrictions of the monitoring campaign, please indicate how the authors dealt with inadequate data for specific variables. 

The following paper should have been referenced and discussed in the manuscript: 

Grange, Stuart K., and David C. Carslaw. "Using meteorological normalisation to detect interventions in air quality time series." Science of the Total Environment 653 (2019): 578-588.

Yang, Jiani, Yifan Wen, Yuan Wang, Shaojun Zhang, Joseph P. Pinto, Elyse A. Pennington, Zhou Wang et al. "From COVID-19 to future electrification: Assessing traffic impacts on air quality by a machine-learning model." Proceedings of the National Academy of Sciences 118, no. 26 (2021).

Vu, Tuan V., Zongbo Shi, Jing Cheng, Qiang Zhang, Kebin He, Shuxiao Wang, and Roy M. Harrison. "Assessing the impact of clean air action on air quality trends in Beijing using a machine learning technique." Atmospheric Chemistry and Physics 19, no. 17 (2019): 11303-11314.

Le, Tianhao, Yuan Wang, Lang Liu, Jiani Yang, Yuk L. Yung, Guohui Li, and John H. Seinfeld. "Unexpected air pollution with marked emission reductions during the COVID-19 outbreak in China." Science 369, no. 6504 (2020): 702-706.

The clarity and context need significant improvement to better draw out why the results are significant.

The manuscript is generally well written and clearly presented. However, its research outcome (i.e the impact of meteorology and regional sources on air quality in Buenos Aires, Argentina) is not new. It should investigate the interactions between input variables to understand more about Random forest model. I do recommend publishing this work if the authors can solve my major concerns as below:

Major concerns:

1.       Selection of explanatory variables:

1.1. Table 3, line 200-205: Air quality strongly depends upon boundary layer height and long-range transport. Why were these variables not included in this study as input variables in the model? Please refer to a reference by Shi et al. 2021 (Science Advance, Vol 7, Issue 3, “Arupt but smaller than expected changes in surface air quality attributed to Covid-19 lockdowns”).

1.2. Could the author explain why the cho CO, NO as explanatory variables for NO2? CO and NO were modelled from t2, rh2, U, V and gasoline diurnal patterns, so I guess the author also can model NO2 based on these variables. Similar questions for explanatory variables for SO2 and PM10, and O3.

1.3.  In the model of NO2, did the author investigate interactions between input variables such as NO with t2.

1.4.  In terms of O3, it strongly depends upon atmospheric temperature. Why does this variable not be included in your model?

2.       Testing dataset:

2.1. Figure 2, Line 189: What criteria do authors select testing dataset based on (i,e 2 weeks data before lockdown?)

2.2. In my opinion, the 2-weeks data for testing data sets is too short. Therefore, authors should do a model performance for at least one month before and after the lockdown/partial periods.

 Minor comments:

3.1. Table 4: I think author should include the r value between model and observation rather the r-value for diurnal cycle (r-dc)

3.2.  Table 5: In BLD, it should include concentrations of pollutants between observation and model.

3.3.  In discussion: Authors should plot the dependence of concentration of pollutants on meteorological conditions.