Comment on “Estimation of the vertical distribution of particle matter (PM2.5) concentration and its transport flux from lidar measurements based on machine learning algorithms” by Y. Ma et al.

Vertical distribution of PM2.5 and its flux are of great importance to evaluate its impacts on environment, climate as well as human health. The article proposes a new method for estimating vertical distribution of PM2.5 concentration from active remote sensing observation based on ML algorithms. The topic is of sufficient interest to the communities of study of atmospheric aerosol and environments. In general, I find this manuscript to be of interest for publication and appropriate for Atmospheric Chemistry and Physics. There are several suggestions for improvement listed below that should be considered by the authors and the editors before publication.

1. P3, line 17, the abbreviation of “unmanned aerial vehicle” should be added.
2. P6, Line 7, is the R2 the correlation coefficient or determination coefficient? please confirm it.
3. P8, Line 16, please clarify the level of significance test.
4. P9, Line 8-9, “we randomly pick 90% (4,807) as a training dataset, and the remaining 10% (535) as the testing dataset.” I am confused the method of picking the training and testing dataset, please focused two questions: 1）After multiply and randomly picking samples, do the so-called remaining 10% samples participate in the model training? Or are the 10% not involved in the training of the model at all times? If the first one, this means that the predictive performance these models are unreliable. Please give the detailed explanation.
5. About the validation of model training in section 3.3 and the evaluation of predictive power in section 4.1, the authors should consider more methods, e.g. sample-based 10-fold cross-validation.
6. P12, Formula (5), please give the unit of transport flux in the corresponding context so as to understand it conveniently, because the unit of transport flux at a certain height is different from that of column-integrated transport flux.
7. P13, Line 24, model à models
8. Section 5 should be rewritten. This section is just repeating some statements that have been made in the previous sections. In a good conclusion, the authors should interpret all the findings and even discussion with a higher level of abstraction.

In this paper, authors want to Estimation of the vertical distribution of particle matter (PM2.5) concentration based on machine learning algorithms, that's a good idea. This is a problem worth studying. Before publication, there are some problems to pay attention to.