I really enjoyed reading the manuscript, that is presenting an innovative methodology to classify drizzle presence in the clouds using a machine learning algorithm. The presented results are robust and potentially open new future studies and analyses. I find the paper well written and I only have some small minor comments regarding the exposition of some scientific concepts. Despite the completeness of the exposition, I could find a couple of open questions regarding the machine learning algorithm and the physical interpretation of the classification, that I would like to pose to the authors, to possibly improve the manuscript further. Please find below the detailed list of my comments.

Best regards

General comments:

This a very important study that will have a significant impact on the field. The article is well structured and written. I have no major comments, just a few minor suggestions. I would like to propose that the article is accepted after the minor changes.

My main suggestion is to use standard verification metrics when presenting the performance of the method, i.e. In Fig. 2. I would suggest that you show probability of detection, false alarm rate and critical success index as functions of Z, instead of CDF.

Minor comments:

Line 157: “…cloud/drizzle datasets is trained by a machine learning algorithm…”

Do you mean a machine learning algorithm is trained by the datasets?

Line 170: “…turbulence broadening is set as 0.2 m/s which is obtained from local observations…”

You use spectra width for cases where Z is less than -20 dBZ, do you do any other data filtering? As you show later, spectra shape might be modified by auto conversion even for such low reflectivity values.