This is an interesting paper for exploring and discussing new possibilities to perform atmospheric trace gas measurements using satellite-based spectrographs that try to fundamentally improve on the concepts of predecessor instruments.

Traditionally, the sizing of predecessor instrument has been driven by user requirements on signal-to-noise and spectral resolution and sampling that are in some cases questionable when the resulting L2 data products are inspected afterwards.

The paper could benefit from a section of requirements in order to clarify which L2 data products are targeted with the optimised spectrometer design.

For example, from use perspective, the high spatial resolution of e.g. 1 km x 1 km in the spectral range 270-500 nm would 'only' benefit NO2 retrievals, for which a very limited spectral range (e.g. 420-450 nm) with high spectral resolution would be sufficient.

It is the combination of various requirements for various L2 products that are all given priority 1 that often drive the size and mass of these type of satellite UV-VIS-NIR-SWIR spectrometers.

Properly accounting for polarisation has had a tendency in the past to increase instrument size (see also section 3.5).

Accepting the resulting errors from ignoring polarisation can be used to reduce the instrument size (and mass) considerably.

In my view the paper could be further improved with some additional comparison information:

• Size comparison for predecessor instruments.
• Size comparison for 1D vs 2D predecessor instruments.

Section 3.5 Further considerations

I don't share this statement:

"Obviously, for very small spectrometers the depolarizer will also be very small, thus  adding negligibly to the volume and weight of the instrument."

Depolarizer plates work on the basis of spatial randomisation of the atmospheric polarization, which requires a certain minimal spatial size of the entrance aperture and depolarization plates. Has this been considered?

Section 3.6 How to combine the signal of a large number of spectrographs?

This chapter is a bit an oversimplification of a post-processing issue to combine the data of the various spectrometers that may become problematic in terms of efforts and processing power.

Section 3.7 How to manufacture arrays of (micro) spectrographs?

This is a bit of an oversimplification, because the issue is not so much the manufacturing costs, but more the space qualification and documentation costs.

Section 4.1 is not fully understood, because here the authors start to use two competing objectives without making clear (to the opinion of the reviewer) how these are combined:

1. Using multiple small spectrographs to each observe a different ground pixel, whereas an instrument such as TROPOMI observes all ground pixels simultaneously.
2. Using multiple small spectrographs to each observe the same ground pixel, in order to improve signal and signal to noise using multiple similar small spectrographs instead of one bigger one.

For example, in table 2, the TROPOMI type has 576 ground pixels per spectrograph, whereas the 'Scaled 1' has 6 ground pixels per spectrograph.

Hence to cover the same spatial range and resolution on the ground the 'Scaled 1' needs 576/6=96 spectrographs.

The number of spectrographs difference is 100, which thus covers only for the above spatial range/resolution per spectrograph, not for signal / signal-to-noise.

Using mechanical scanners, as mentioned earlier in the section, doesn't comply with the simplicity and small design of the downscaled spectrographs.

I find the conclusion section 5.2 somewhat biased and not considering all advantages and disadvantages equally, and to some extent also comparing different things.

See also the above comments.

I am not convinced that at the same spatial resolution and range, same spectral resolution and range and the same user / L2 requirements the option of the array of

identical downscaled spectrographs presents a significant mass volume advantage. I find this aspect is not conclusively demonstrated in the preceding sections and I encourage the authors to improve this demonstration, including the aspects also mentioned here, e.g. by clearly separating the aspects of number of ground pixels per spectrograph and signal/signal-to-noise per spectrograph.

I also recommend that clear recommendations for the downscaled spectrographs for spectral range/resolution and the use of two-dimensional detector arrays vs the use of mechanical scanners are given, because the use of scanners in small spectrographs is not an option and the use of two-dimensional arrays complicates the optics (as pointed out in the paper).

In addition, operating a fleet of small satellites with small spectrographs is an expensive undertaking. Which operational institution (or company) is supposed to do this and at what cost?

That having been said, the paper contains a number of interesting and valuable design points that certainly deserve further discussion.

To guide that discussion better I recommend to account for the above suggestions and questions.