Review of UFS-RAQMS Global Atmospheric Composition Model: TROPOMI CO Column Assimilation by Maggie Bruckner and co-authors.

The paper presents the development of a whole new modelling system by coupling chemistry and aerosols representation to NOAA’s UFS, in order to assimilate TROPOMI CO.

I am just having a hard time following the discussion about aerosols. You mention that sulfate aerosol formation will change with OH. Does the CO assimilation impacts aerosol formation and are the simulated AOD different between the control and the TROPOMI DA ? What about secondary organic aerosols and heterogeneous chemistry ? This is important to clarify as these should be pathways AOD would change following CO assimilation (without optimizing emissions or transport). Specifically, if AOD are not changed following TROPOMI CO, does it mean that the aerosols are well simulated but not CO (on figure 13) ?

Regarding Figure 14, I don’t really follow the practice of increasing AOD by a factor of 3, why don’t you just scale the emissions of biomass burning CO and aerosols, using the information from the CO assimilation, even a rough estimate and run forward simulations ? Since you are using the same biomass burning emission inputs, AOD and CO errors must be correlated. 

Similarly, with the field campaigns, you have the opportunity to evaluate the impact on other species and assess potential improvements.

The FTIR observations from NDACC uses a different remote sensing approach than TROPOMI to retrieve CO, but I think it is somehow misleading to call them profiles without caution. I would mention the degrees of freedom for signal, which are probably high enough to allow for a retrieval of a partial tropospheric column at best. If you are willing to keep the figures with the profiles, I would also show the NDACC prior and add appropriate disclaimers. 

Minor comments:

L33 please define the TROPOMI acronym and reference here (from line 137: Tropospheric Monitoring Instrument (TROPOMI) (Veefkind et al., 2012)).

L139: define the acronyms UV-near IR and shortwave IR 

L145 (paragraph): Super observations are used to match the model grid’s spacing, to reduce noise and avoid overfitting observations, as well as for computational efficiency. Sekiya et al., (2021) found that the smoothing of the spatial variability in analysis increments with OMI NO2 assimilation and noted that the super observations were reducing the number of assimilated observations by a factor of 10, making the use of super-observation to be more relevant with TROPOMI. Note also that the smoothing is actually a desired effect by reducing noise and removing the sub-grid variability that the model cannot represent. I understand there might be a loss of signal while producing the super-observations, so it would have been desirable to actually perform the experiment. You would have been able to show whether the statement line 150 (an underestimates in localized CO column enhancements) ends up being correct or not.

L253 and L275: the precision of the Picarro instrument is about 1 ppb for 1-min averages. So, it would be more appropriate to round the biases to significant figures of 1 ppb or 0.1 ppb

Figure 5: MOPITT retrievals include the dry-air atmospheric column, and it is making it easy to convert both model and observations to XCO. You could also show the bias for Control and TROPOMI DA on panels e) and f).

General comments:

The manuscript “UFS-RAQMS Global Atmospheric Composition Model: TROPOMI CO Column Assimilation” written by Maggie Bruckner et al. develops the framework of TROPOMI CO data assimilation with the UFS-RAQMS model and 3D-Var, and evaluates its performance against independent satellite, aircraft-campaign, and ground-based remote sensing observations. The manuscript demonstrates the advantages of assimilating TROPOMI CO over the model simulation. This manuscript is generally well written, organized, and designed. I recommend inviting the authors to revise their manuscript to address specific points before a final decision is reached. I provide several comments that need to be addressed before moving forward publication process below:

1. Authors newly construct the background error covariance (BEC) matrix by blending the NMC method and the sensitivity simulation with reduced biomass burning emissions. I would suggest adding sensitivity calculation using (1) blended BEC, (2) NMC BEC, and (3) BB emission BEC. It would be helpful to demonstrate and understand the advantage of the blended BEC approach for readers.
2. Authors state that the UFS-RAQMS experiments include assimilation of MODIS AOD in the methodology section. However, the experiments largely underestimate the VIIRS AOD by a factor of 3. What is the reason why such large negative biases remained even though AOD was assimilated? Also, supposing that these negative biases are attributed to biomass burning emissions, it is more appropriate to apply a scaling factor of 3 to the biomass burning emissions than to the UFS-RAQMS AOD fields.

Specific comments:

p. 1, l. 24—25: What implications the improvements in biomass burning AOD/CO relationships by TROPOMI CO assimilation provide?

p. 2, l. 44 “CTM forecast vary …”: Please clarify if this sentence means the CTM CO levels vary or the CTM forecast performance vary.

p. 2, l. 45-48: Is the scheme with the ratio relative to CO for determining the release of other species commonly used? If so, please cite the relevant literature.

p. 2, l. 51 “CTM fields”: Does it mean concentration fields?

p. 2, l. 52–54: Chemical DA methods include ensemble Kalman filter approach in addition to the methods authors mentioned.

p. 2, l. 61–62: Please explicitly state what is the advantage of applying TROPOMI CO DA to UFS-RAQMS over previous studies and/or previous version of RAQMS.

p. 4, l. 114–116: What product version do authors used for MODIS AOD, OMI total ozone column, and MLS ozone profiles?

p. 5, l. 131–135: How did authors determine the inflation factor for standard deviation and the model levels at which two BEC estimates are blended?

p. 5, l.147–149: How do authors account for spatial representativeness errors in the observation errors during the analysis steps?

p. 7, l. 181–183: It might be helpful to show analysis increments for discussion on transport impacts after the data assimilation.

p. 11, l. 247: Please clarify how to define in-plume measurements.

p. 13 and 15, Figures 8 and 10: How does the blended BEC approach affect the improvements in CO vertical profiles?

p. 16, l. 449—450 “We only assimilated the profiles at 700 hPa and the vertical localization reduces the impact towards the surface.”: It should be described in the method section.

p. 16, l. 490—493: What implication does this contrast btw 500 and 700 hPa provide for model and observation errors?

p. 16, l. 495—496: What about posterior emissions from MOPITT-DA? I'm curious about CrIS impacts on emissions. Could you show it in the main text or supporting information?

p. 16, Table 1: I would suggest adding statistics such as mean bias to Table 1 for readability.