This paper used the random forest models to predict typhoon-associated air quality in the Guangdong-HongKong-Macao Center Bay Area. This work is interesting and meaningful to the readers of AMT. Generally this paper is well organized. The methods and conclusions are reliable. But I still have some suggestions below.

Detailed comments:

1. This paper used data from 36 air quality monitoring stations in 10 cities of the GBA that is heavily polluted. Why did you not consider the observation data from the rural regions of Guangdong Province?

Are the 36 air quality monitoring data are used for both model training and RF model evaluation? I think it may be more reasonable if you could choose parts of the 36 sites for model training and the others for RF model evaluation.

2. In section 3.1, did you compare the model performance of the RF model with the other traditional air quality models, e.g., CMAQ, WRF/Chem.

Figure 5: Why the data are only available over the seas?

Figure 6: The same data at the air quality monitoring sites are first used for model training and then for model evaluation?

This study employs  the random forest models to predict typhoon-associated air quality quantitatively in the Guangdong-HongKong-Macao Center Bay Area. The prediction models are estabilished for typhoon and non-typhoon days. Thus, the results suggest that different air pollution control strategies for typhoon days and non-typhoon days should be adopted. The work is innovative well written and interesting to the readers of AMT.

I have two questions below.

1) The present study takes 36 air quality monitoring stations in 10 cities in the GBA (Guangzhou, Shenzhen, Zhuhai, Foshan,

Zhaoqin, Jiangmen, Huizhou, Dongguan, Zhongshan, Hong Kong) as research objects. Why did you not consider data from the rural regions? 

2) The study used ERA5 reanalysis from meteorological data. Couldn't these data be integrated with those coming from other high-resolution instruments? For example lidar or meteorological radosondes?