The detailed comments, please see the supplement!

This study implements Voronoi and spherical ice crystal models in observed brightness temperatures at 380, 640 and 874 GHz to retrieve Ice water path(IWP) and effective particle radius (Re). Authors show that Voronoi model can better reproduce results compared to previous 'standard' values. I think such result is obvious and can be easily predicated as simple spherical model is less adequate for irregular-shaped ice particles, but authors showed it quantitatively. In my opinion, there exist mistakes/incompleteness in English language, methodology, and scientific discussions, as outlined in specific comments below. Authors are suggested to improve the manuscript by considering the comments given below as well as by carefully improving the English writing/expression.

1. Line 19: '.. we completed the..'--> completed is what sense? rewrite it.
2. Line 48 : '..signfinifiantly developed...': rewrite the sentence.
3. Line 132: What is 'CoSSIR-MCBI' algorithm? A brief description is important with relevant references.
4. Lines 140-143: Specify ice particle sizes either in table or describe size interval in the text. Provide same information for 20 wavelengths mentioned here.
5. Line 149: Specify what refractive index values are used for ice particles here. Give a reference.
6. I think Figure 1 is confusing. Either make Figure 1 clear or remove it and describe the mythology clearly in the text.
7. Line 179: I do not understand how clear sky days are selected here. Are they before and/or after the cloudy sky days or average of certain week (or month etc.)?
8. Lines 188-189: 'statistical multiple linear regression method': Write a few lines to clarify it. For example, what are dependent and independent variables in this method?
9. Line 199: A reference for Eq. (3) is required.
10. Lines214-219: Please specify the coefficient terms either in table or in text.
11. L222: Is BTD_1-3 is same to BTD_1-2-BTD_2-3 ? If so, better to write BTD_1-3 instead of BTD_1-2-BTD_2-3.
12. Lines 227-229: I think BTD depends strongly on cloud top temperature as well as surface temperature along with cloud properties. Since they are fixed here, errors are expected in retrieved values. Can authors provide error ranges in retrieved parameters due to such assumptions? If possible, authors are suggested to use actual data from cloud top and surface temperatures rather than assumptions.
13. A theoretical perspective should be given for using BTD_2-3 and BTD_1-2 -BTD_2-3 in Eq. 4. (Why  these two are important in cloud properties retrieval among several possible combinations of three wavelengths). Further, there should be an error term in Eq. 13. It is further necessary to describe in detail about the methodology. For example, what are the convergence criteria, how the initial values are determined, and how measurement errors can affect the retrieved values etc.
14. Line 250: Since the x-axis is size parameter (not radius), it is difficult to understand where 120um exist. Either rewrite the text or make Figure 2 clear by adding additional x-axis.
15. Lines 250-254: Why large difference exists for large sized particles in Figure 2 remains undiscussed.
16. Lines 261-264: What could be the plausible reasons for such results for relatively larger particles?
17. I do not understand why BTD2 is shown in Figure 6 as it is not used in the retrieval (see Eq. 4).
18. Please make Figure 7 easy to understand. For example, indicate the values of IWP and Re with dots ( e.g., Nakajima and King plot). As there are overlapping lines in Figure 7, how cloud properties are retrieved if data fall under such overlapping lines? 

The paper assessed the capability of the Voronoi and sphere models in the retrieval of IWP and r_e using aircraft-based measurements of 380, 640, and 874 GHz brightness temperature. Based on the sensitivity analysis, the brightness temperature differences between 640 and 874 GHz are used for IWP retrieval, while brightness temperature differences between 380, 640 and 874 GHz are used for r_e retrieval. The authors find well correlations between the Voronoi-based retrievals and Evan’s Bayesian retrievals using data from the CoSSI instrument. The comparisons of the retrieved IWP and r_e between Evan’ s Bayesian retrievals using data from the CoSSIR instrument and the inversion algorithm among the Voronoi and Sphere models suggest that the Voronoi model outperforms the Sphere model. Overall, the highlight of this paper is that the Voronoi model has been previously applied to visible and infrared applications in satellite remote sensing and climate model simulations, and now it is being applied over the terahertz region to investigate how well the model performs there. The paper is relatively well-written, and the figures are also well-displayed. The analysis is quantitative and clear, with no obvious flaws. This paper could be a good supplement to the development of satellite remote sensing of ice clouds in the sub-millimetre regions. The topic presented in this study is suitable for Atmospheric Measurement Techniques, so I recommend Minor Revisions for publication.

Specific comments

1. To ensure the effectiveness and representativeness of the Voronoi model in terahertz region, I recommend the authors compare the retrievals against more other ice crystal scattering models, such as the column aggregate?

2. How about the accuracy of retrieved IWP and r_e in previous research in terahertz band? Please illustrate in the introduction.

3. Line 98: The “GOA” acronym has not been defined.

4. I recommend the authors add more analysis and possible explanations in the result section. I recommend the authors relate the single-scattering results to the retrieval results.

5. In Figure 9, the scattered dots are hard to statistically measure the accuracy. Figure 9 might be better plotted as a PDF of the retrievals.

6. Line 117: “Mo.,del” should be “Model”.

Please find the attached comments.