This work proposes a method to estimate cloud envelope and cloud development velocities based on high resolution satellite train imagery. Because the planned satellite train is not launched yet, the work relies on simulated image data to show the proof of concept. Although the simulated data are perfect in the sense that noise is not accounted for, they are helpful to compare against the ground truth (that will not be available in the case of real data) and therefore show the potential.  Before publication, I suggest thatthe following points are addressed:

Line 12-14: “An independent method based on optimizing the superposition of the cloud top, issued from the atmospheric research model, allows to obtain a ground estimate for the velocity from two consecutive acquisitions.” I do not understand this statement, what is meant by superposition in this context? Maybe it is more clear in the text but it just makes the abstract confusing.

Similar for the lines 14-17, please considering restating these lines to make it easier for the reader to understand.

Line 56, Do these error bands differ by magnitude of the velocity and the height? If so, is it better to give a percentage?

Line 75, Please be specific, magnitude acceleration in what? Computation time?

Line 101, Do I understand correctly that the image will not be taking images continuously but will start when triggered and for 200 seconds only. If so, please make this explicit and explain what will trigger the image capture event.

Line 110, How accurately can the satellite positions can be retained? That is what are the bounds of change in baseline? Is this likely to impact the retrieval quality?

Line 140, is 22.5 km the location from a reference point? Please clarify.

Why do Shallow cumulus clouds have better resolution? Shouldn’t it be the opposite?

Lines 327-333, is there a specific reason for merging 2xtwo-view  instead of using three-view when data from Sats 1-2-3 are used?

Figure 10,  Do I understand correctly from the figure that no points are retrieved with Sats 1-3 and Sats 1-2-3 scenarios in A3-5 and A8-9 views? If so, how can you say that none of the configurations outperform? Also, what is thee reason for the skewed-towards-A9/A10-views distribution of error in z in Figure 10.b?

Equation 10, why are the subscripts suddenly 9 and 11?

Line 418, please explain why a dual mode caused by a possible “divergence of the cloud top in the center right part”  would not show in the GE distribution?

This paper describes a method to estimate the 3D cloud envelope and development velocity using simulated images of a triplet of small satellites in a sun-synchronous orbit at 600 km height. The focus lies on trade wind cumulus and deep convection, while the methodology relies on stereo analysis and tracking to compute the 3D points of the cloud envelope and subsequent cloud motion estimation. The study assesses the feasibility and accuracy of the proposed mission design.

The paper is well written and structured and the study is thoroughly conducted and the topic is scientifically relevant. I suggest publication after the following points have been addressed:

1. Fig. 1:

Could be a bit more detailed. Just a few more numbers and lines if possible. For example the height of and the distance between the individual satellites might be good for directly understanding the geometric setup, which is important for stereo analysis and triangulation.

2. Line 106:

“consequently simultaneously”. Sounds a bit strange. Maybe just leave the “consequently” out.

3. Was it described somewhere what axis the along-track direction was (x or y)? Maybe add it to one of the initial figures so that analysis later is easier.

4. Line 115 + 116:

For a tilted view shouldn’t the track resolution be decreasing as a larger area is projected to a smaller image area? The same with the Ground Sampling Distance. Shouldn’t it increase for a tilted view? Maybe just a misunderstanding.

5. Line 185:

I think the field of view is always constant as it depends on the camera. If the angle representing the image projection of the ground area is meant then it should be smaller with tilted view, shouldn’t it?

6. Line 278:

You write that the cameras are affine for a small tile of an image. Considering the field of view in this simulation of 1 degree, the cameras already are very weakly perspective. Does an additional tiling matter?

7. Line 279:

Maybe shortly describe what an epipolar line is and why it’s useful.

8. Line 279 / Fig. 8:

You describe that you conduct a stereo image rectification in order to make the stereo analysis easier, aren’t you? In that case the y-component is usually zero (which you write in line 285). Is that correct? Also the disparity (parallax) should have values between 0 and infinity (or negative). But in Fig. disparities are both negative and positive. A negative disparity would mean that the observed point is behind the cameras. Or do I miss something?

9. Fig. 10:

What explains the large differences in Fig. 10F for A10 and A11? Shouldn’t the differences at least be symmetrical / similar to A1/A2? Similarly Fig. 10b. What could be a reason?

10. Sec. 5.3, comment: It is good that you mention possible differences due to the different distance estimation methods.

11. Fig. 12:

A sigma of 22.85 m/s in Fig. 12a for Vz seems a bit large considering the histogram.

12. Line 443:

“attitude” → “altitude”?