In this paper the authors present a comparison of CO TROPOMI with ACE-FTS in the Arctic. To provide context to this comparison, TROPOMI and ACE-FTS CO products are compared against NDACC measurements from PEARL-FTS in Eureka. The study suggests that TROPOMI CO product exhibits a high bias in the high-Arctic region.

This study has been thoroughly conducted. The paper is well written and interesting. I found the paper clearly presented and well organized. This work is useful for the TROPOMI team to improve the CO product, as well as for the community.

The subject of the paper is appropriate to AMT. I recommend the paper to be published, after addressing the minor issues I raise below.

Comments:

L50: add TES +reference

L85: methodology used for comparing each instrument IS described

L221: add units of CO reference profiles

L265: null-space error e_n (also known as the smoothing error). Why not prefer the smoothing error ? You should add the Rodgers reference. If you want to use the null-space error, you have to explain why. You give references of different studies but we need explanations.

L285: In the following section describes.. Remove “In”

L300: A is the VMR averaging kernel > A is the VMR/VMR averaging kernel

 (A is unit less)

Same thing in L343

L328-330: What reference should you give when talking about sensitivity ? Rodgers 2000 ? Rodgers and Connor 2003 ? As already said, you mention a work published in a paper “following xx et al.”. You should be more specific and cite the proper reference for terminology.

L362: SEM. Define (to understand the difference with standard error and standard deviation) and justify your choice for using SEM. Make uniform notation.

Caption of Table 3: “standard deviations of the differences” in the first sentence. “standard error of the mean” in the last sentence. It is confusing. The caption should be reformulated.

L481: coming from the measurement IN this range ?

L487: “do not overlap within their bounds of uncertainty on the standard error” Not clear, please reformulate

The paper provides an interesting and valuable addition to the already published validation papers on the S5P CO product. 

Comments/questions:

p3, l63: the destriped columns are now available (since processor >02)

§2.1: suggest to mention the definition of the tropomi grid  

p4, l110: "relatively insensitive" ... a bit a confusing phrasing to me, I would say a column avk of 1 means all layers contribute equally?

p4, l120: as mentioned in the product readme file (PRF) §2 the processors before 010202 contained flaws (eg wrongly flagged sunglint pixels). Actually from https://s5pexp.copernicus.eu by combining RPRO and OFFL you can skip any processor below 010202 (see latest report on https://mpc-vdaf.tropomi.eu/ or the PRF on https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-5p/products-algorithms)

§2.2, 2.3: suggest to add uncertainty estimates (similar to §2.1)

p6, l164-165: suggest to use www.ndacc.org and add cams27.aeronomie.be to specify what is meant with the CAMS rapid delivery service

§2: general comment: add more information on measurement characteristics: plots of a typical avk, typical uncertainties, typical dofs, … this is important to understand some of the arguments further down

p7, l205: can you give some information on the typical number of tropomi pixels in a co-location (to know how the threshold of 50 relates to the typical number)? 

p8, l215: can you explain what you mean with pixel-to-pixel biases (vs pixel-to-pixel variability)?

p8, eq 1: depending on the VMR profile x being with respect to dry or wet air, this equation will alter. Can you add this specification for x and add a reference to the definition of the column averaged g? The equation is derived from the hydrostatic balance equation and should be (for x being a vmr profile wrt dryair)

\[

\Sigma^N_{i=1} \frac{\Delta p_i (1-q_i) x_i}{M_{\mbox{da}} g(h_i,\lambda)} 

\] 

(so no column averaged g but the altitude dependent g and using a specific humidity profile q). Did you approximate q ~ 0, can you motivate this? Why using the column averaged g instead of g from WGS84 eg?    

p9, eq 2: same question: x being with respect to dryair will alter the equation and an approximation is used? 

p9, l237: can you explain why the reference profile is used and not the “retrieved profile” (=scaled reference profile)? 

p10, l269: I assume the true profile in eq 4 is the (unknown) true profile, while in eq 3 it is the ACE-FTS profile? If so, I would suggest to use another label for ACE-FTS in eq 3 (and where applicable).

p10, l285: “. In the following section describes…” -> “The following…”

p12, l328: sensitivity depends on the AVK units: did you use the AVK acting on vmr profiles (as reported in the GEOMS NDACC files) or the converted AVK acting on vmr profiles relative to the prior. The latter should be used for the sensitivity (SFIT4 uses relative units).

p14, fig2: the figure labels (a,b…) doe not correspond to the labels in the legend. The figure (f) has a different scale on the x-axis (1e17) compared to the y-axis (1e18).

p18, l423: the sign of the null-space also depends on the sign of (I-A)? I would need a bit more information to understand the statement..

p21, fig8: suggestion to use the same scales on the x axis for both subfigures: this will allow to see the effect of the smoothing more clearly 

p22, l479: “mean total column DOFS” -> “mean total DOFS”

p25, l 512: ‘latitudinal dependence … present in both the unsmoothed and smoothed cases’:  can you relate this latitudinal dependence to the reported uncertainties? Is it significant?

Thank you for your replies.