

Reviewer 1

Reviewer tjqh, Strength And Weaknesses: Strength: 1. The paper addresses an important problem in air pollutant concentration prediction. By integrating the mechanical model and machine learning, the work bridges the traditional methods and current trend. 2. The high-level presentation is good which is easy to follow and understand. 3. The experiment is solid and comprehensive.

annex

Author response: We are very grateful for your comments and recognition of the strengths of the manuscript.

Weakness:

1. The paper uses inconsistent terms and symbols, which is confusing at some time For example, R and r_a are to represent the region in section 3.1, but authors use the same symbol for messages in equation 2, which is a little ambiguous. In figure2, the input data doesn't correspond to the variables in the definition section, I would suggest the authors to use consistent symbolic notations for the important features in the both the text and the explanatory figures. Some of the figures need more explanation, e.g., in figure1, the meanings of the axes are unknown. Typo: neighbor node net->neighbor node set 2 The novelty is not outstanding. Although integrating the mechanical model and machine learning is interesting, most of the techniques are well-established and the authors simply apply them to the new problem. The overall framework is the same as this paper: Jiahui Xu, Ling Chen, Mingqi Lv, Chaoqun Zhan, Sanjian Chen, and Jian Chang. HighAir: A hierarchical graph neural network-based air quality forecasting method. arXiv preprint arXiv: 2101.04264, 2021.

Author response:

1. We thank the reviewer for this comment. We have revised the issues of inconsistent terms and symbols, figures explanation, typo and so on. We believe that the quality and readability of the revised manuscript have been significantly improved thanks to these valuable comments.

2.. The main contributions of this paper include:

• We innovatively propose a dynamic spatiotemporal graph model combining mechanism model and graph neural network. The adjacency matrix and edge weight vector of dynamic graph are constructed based on the simulation results of diffusion, transport and deposition of polluted air mass by mechanism model, so that the

