The manuscript by Seo et al. describes the application of the DOAS algorithm to the GEMS visible spectra to retrieve tropospheric NO2 vertical columns. Compared to the current (v2.0) operational NO2 retrieval from GEMS, The DOAS implementation by Seo et al. included significant improvements in slant NO2 column estimations and their conversions into vertical columns. This manuscript is well-written, and I recommend its publication. I have a few comments that the authors may address to help a reader better appreciate the algorithm implementation.

1) The authors used a broader spectral range (425 - 480 nm) to improve the slant column estimation, particularly in reducing its noise level compared to that from the narrower range (432 - 450 nm) used by the operational algorithm. However, the manuscript contains little discussion about the slant column biases. How do the biases from the broader window compare with those of the narrower window of the operational GEMS NO2? Are the biases higher, lower, or similar in magnitudes and north-south behavior?

2) The broader spectral range (425 - 480 nm) includes possible soil signatures (Richter et al., 2011). Does slant column fitting include a soil signature term over areas where it may be present?

3) Eq. (1) (page 4) contains the offset term (offset(λ), a linear function of wavelength λ). Please describe the impact on the slant column estimation. Does its inclusion reduce the noise level of the slant column or change (consistently increase or decrease) the slant columns to reduce biases? In short, please justify this offset term.

4) On Page 5, line 141, a pseudo-cross-section for polarization correction is added to the slant column fitting. Please describe its impact on the slant column.

5) On Page 10, line 267, a low-order bivariate polynomial is mentioned. Figure 5 suggests that it is only a single variable (i.e., latitude) polynomial.  Looks like the UTC is a label only. Since the polynomial is time-dependent, the polynomial should indeed be bivariate, with the second variable being the longitude. However, the time difference is probably sufficiently small that the longitudinal dependence may be neglected.

6)  The authors used the OCRA cloud fraction in AMF calculations, avoiding the cloud fractions from the GEMS standard product, which likely biased high due to high (likely ~8% or higher) biases in GEMS sun-normalized radiances. However, the authors selected GEMS (v2.0) cloud pressure and GEMS (v2.0) background surface reflectance (BSR) for AMF calculations. Please comment on the possible biases in these products due to radiance biases.

The authors describe a new algorithm to retrieve NO₂ from the GEMS satellite instrument. The algorithm is based on what has been used for polar-orbiting satellites and has been modified for application to geostationary satellites. The authors evaluate the retrieved NO₂ columns using TROPOMI and the operational GEMS retrievals and discuss the uncertainties in the retrievals.

This is a high-quality manuscript. It includes a complete description of the new algorithm and a thorough analysis of its strengths and weaknesses. It is well-written and makes good use of figures. I point out below a few points that could be better addressed.

1. The reader would benefit from a brief description of the operational GEMS algorithms near the beginning of the paper. It would help to include a table describing the major similarities and differences between the new and the operational GEMS algorithm.

2. The uncertainty analysis (Section 3.4) could be improved by discussing uncertainties specific to retrievals from geostationary satellites. For example, how do the retrieval uncertainties vary with the time of day, or how different are they near the edge of the field of view compared to the center?

3. The effect of aerosols on the NO₂ retrievals is important for Asia, and in most retrievals, it is considered implicitly in the cloud parameters retrieved using O2-O2 absorption. It is unclear whether the OCRA algorithm used in this work does the same. Does it instead correct for the presence of aerosols and could that partly explain why it retrieves lower cloud fractions (figures 9 and 10) compared to the GEMS retrieval?

4. Equation 6: Is there an error correlation between albedo and cloud fraction (Boersma et al. 2018; doi: 10.5194/amt-11-6651-2018)?

5. Line 547-9: NO₂ is also photolyzed by visible radiation, not just UV. Another factor for the low noontime values of NO₂ is oxidation by OH.