Please see the attached pdf.

The authors presented a PWV vertical correction grid model using the Random Forest and ERA5 monthly average hourly data. The performance of the new model was evaluated with ERA5 and radiosonde pressure levels PWV and the newly developed PWV vertical correction model (C-PWVC1) in aspects of seasonal and spatial locations and height differences. The results denoted that the new model is superior to C-PWVC1 across the research regions. Overall, the paper is written well and logically. Therefore, a minor revision is recommended for this manuscript before acceptance. However, to improve the quality of the paper, the following comments should be solved.

1. L29-30: The discussion of “inconsistent pressure levels (heights) for storing PWV data from different sources hinder the fusion and reliability analysis of PWV multi-source data.” is not convincing enough because lack of approved references.
2. Why Random Forest (RF) is employed for modeling, but not the commonly used Backpropagation (BP) neural network or other machine learning algorithms?
3. L54: The first appearance of "RMSE" should explicitly refer to its full name. Additionally, please review all abbreviations to ensure they are defined at their first appearance (except in the abstract).
4. L89: Please confirm the units for the explanation of g in Eq (2). Please make sure there are no similar errors in the manuscript.
5. L143: What I am interested in is what are the respective 'day of the year (doy)' corresponding to the monthly average hourly dataset for each month when training the model. Please clarify these.
6. L149: 'Subsequently, the target point.'. Rephrase this sentence.
7. L193: We suggest that the color bar range for RMSE in Figure 4 should start from 0. The same is true of Figure 7.

8.L244: What is the output of the four grid point model around the RS station? Please clarify it to help readers repeat your experiments.

1. L257-259: The text about Figure 7b is not consistent with the fact.