I have thoroughly read the article titled “Inferring the controlling factors of ice aggregation from targeted cloud seeding experiments” by Zhang et al.  Overall, I find the article very interesting and of high scientific quality although slightly limited due to experimental range.  The article reports on the quantification of ice particle aggregation rate in cloud with several controlling factors in mind.  They use machine learning to identify the main controlling factors.  Although the study is very novel and the results are of high interest to the scientific community, there is an insurmountable modest dissatisfaction because the experimental space is so limited (the temperature range is limited and liquid droplets are necessarily present in high concentrations). These limitations only moderately take away from the novelty of the experimental design and the quality of the analysis.  Overall, I feel that the article is definitely suitable for publication in ACP with some comments below.

My main concern is about the utility of the results.  While it is spectacular that the results have been so well studied, the authors should consider more closely where the results will be used and in what form it is best for those who will use them.  Aggregation rates are used in high resolution models.  The final results presented, while they may be the most accurate based on the data, produce a moderately rough line.  I feel the article would be far more useful if the authors were to present suggestions as to how the results could be used in a model. 

Overall criteria:

Scientific significance (4 Excellent): The publication tackles a very significant unknown through experiments.  Aggregation rate of ice crystals in a cloud is a very difficult, yet important, number to know.  This publication presents the results of the aggregation rate estimate after an extensive field experiment which attempts to accurately measure the range of possible values in the real world.  The experimental design has been well documented. The analysis of the results (the heart of this publication) stand up in quality to the experimental design.  The reason I don’t give it a 5 is that the significance would be higher if they had been able to calculate aggregation rate in a wider range of temperatures which was likely limited by experimental constraints.

Scientific quality (4 Excellent): Overall, the scientific quality of the research is excellent.  The authors use the latest in Machine Learning techniques to get at the analysis.  See specific comments regarding areas where the scientific quality could be improved.

Presentation quality (4 Excellent):  The manuscript is, for the most part, concise and the figures are well presented and easy to interpret.  English grammar is perfect. The authors have done a good job moving some details into Appendices which makes the article read well.

Specific comments:

Line 10-11:  In many microphysics parameterizations, an uncertainty of 0.08 might be considered “close enough” to 1 to be within the margin of error.  I would suggest presenting a level of confidence here to help the reader understand that you are very confident that it is not 1.0 and random experimental factors drove your mean to be below 1.

Line 19: Don’t forget that ice crystals do grow from vapor in regions that do not include supercooled liquid drops such as in cirrus at -50C where WBF can’t exist. 

Line 100: Since EDR wasn’t an important factor, perhaps it might be easier to reduce the text here and just say that EDR wasn’t an important factor rather than giving the details on how EDR was measured. 

Line 124:  smaller crystals rather than smaller ones. 

Line 146:  Please use terms such as “variability” and “confidence” when possible rather than “uncertainty” (which is interpreted by the general public as “we don’t really know”).

Line 172: While you eliminate EDR as being too important, you might comment on measured EDR versus what would be expected in deep convection.  (The same for other parameters, knowing how your data fit into the zoo of cloud types will help the reader to understand how representative your results are of situations of interest to the reader)

Figure 2:  Are there symbols missing from the top?  There is a blue line, then it gives the Cold temperature range twice, then a red line and warm twice.  I suspect there should be a symbol there but it could be my computer.

Line 240 and paragraph below:  Since there is a significant habit transition between warm and cold, it might be interesting to see if there are enough data points in each of the groups to identify a trend in each.  There might be a natural functional change due to habit that could be hidden by linear analysis.

Line 256:  RR can influence geometry and stickiness if there is still some quasi liquid present?

Line 275:  I am not sure that I see how RR decreases ICNC_t0. 

Line 300:  It would be nice to have a reference for the SHAP calculations.

Figure 10: The obvious question is what happened at -5.3 degrees?  The CatB values all dip there especially the 10^1L^-1 line, the other CatB lines are impacted there as well.  As stated in the initial comments, how is a modeler supposed to incorporate these data into a weather forecasting model?

Lines 491-526:  I think it would be reasonable to add the Geron, 2022 reference on line 491 and not have it on every (but one) of the model algorithms.

Review is attached as a PDF