The study by Souri et al. uses a combination of a box model, CTM output, satellite data and aircraft data to try and estimate ozone production rates in the lower boundary layer. The aim of the paper definitely sits within the remit of ACP and TOAR-II, but I believe the manuscript needs major corrections (though mainly textual) are required before it can be accepted for publication.

Major comments:

1. So, when I read the title and abstract of the paper, it read as if the satellite data was the main dataset/resource used to general the PO3 maps. However, from a detailed read of the manuscript, lots of other data sources are required to achieve the outcome. For instance, the authors use model output from a CTM to derive boundary layer satellite NO2 and HCHO products. This is fine but you have moved a long way from “using satellite data”. However, the bias correction of the TROPOMI HCHO and NO2 using surface column measurements is a good practical step to achieve more robust results. There is then the box-model, which is partial tuned to aircraft observations for some tracers, to evaluate key variables which will go into the final scheme (shown nicely in Figure 2) to derive PO3. Overall, I am happy with the methods used to derive PO3 (especially as the satellite data gives you the high spatial resolution) but I think the actual overarching method of the paper needed rewriting (i.e. instead of putting the emphasis on “using satellite data”, I think you should make it clearer that you use “a synergy of data products” to derive high spatial maps of PO3.
2. In sections 2.2 and 3.1, there are multiple equations but some of the variables are not actually defined, which made it difficult to fully understand the methods without being an expert. So, these sections need to be improved to clearly define and explain what all the variables are in the equations. For instance, on line 161, what are a, b and ε? On line 162, what does i represent? There are also several examples where variables have not been added to the equations (e.g. line 175…I assume these are superscript 2s?). Overall, the method’s presentation needs to be improved and discussed more to make it clear to non-experts what you are using the methods for. There are examples below in the Minor Comments supporting this.

Minor Comments:

Line 100: Define NASA and NOAA in the first instances.

Line 153: “a fixed additive component that is magnitude-independent”, can you provide more detail on what this is and why you use it.

Line 163: Can you give an example of what you mean by “benchmark”.

Line 154: Can you use the satellite column precision as a representation of random errors?

Line 156: Can you be clearer on the text “Moreover, to mitigate this error, its squares are average over a month”.

Line 161/2: Please define what a, b, ε  and i are.

Line 175: Missing variables in boxes.

Line 184: What does BRDF stand for?

Section 2.4: Please provide more information on how you use the MERRA2-GMI data to generate the satellite PBL product?

Line 217: What do n and p represent?

Equation 5 RHS: Should “2” be superscript instead of subscript?

Line 227: Please make it clearer what you mean by “folds”.

Lines 353/354: “Consequently, it is likely that the measurements error resulted in more spread in the comparison”. Can you provide some references to support this statement.

Line 364: Rephase “not to unrealistic” to “reasonable”.

Figure 3: NO2 and NO have the same MB, MAB and RMSE. Is this a duplicate of statistics or coincidence? Also, some of the stats legends overlap (e.g. OH), so the presentation needs to be improved here.

Line 555: One could argue why don’t use just use a CTM or regional model to simulate/output PO3 and supporting variables (e.g. NO2 and HCHO). Would you not benefit from using the satellite and aircraft observations to evaluate the model, identify limitations (e.g. emissions, chemical mechanism etc.), undertake sensitivity experiments to resolve the limitations and then provide more robust estimates of PO3 from the model? That way, you are getting estimates of PO3 but also improving the processed based model providing a better understanding of the processes governing PO3?

Line 560: There are a few instances where you term “significant”. However, do you actually use a statistical test to support these statements?

Figure 11: I might have missed this but do you define “CONUS”?

Figure 17: “The data is based on 2019 TROPOMI observations”. This does not make sense. You list several variables on Lines 652 only which two are actually from TROPOMI. Please update this.

Line 678: I disagree with this statement “satellite-derived product”. As to my major comment #1, you use satellite data, aircraft data, MAX-DOAS data, CTM data, box model data and statistical methods to derive PO3 (as depicted in your Figure 2). Therefore, I believe this needs to be reworded and refocussed (e.g. a data-model fusion approach to derive PO3 etc.).

Main Comments:

The authors develop a powerful parameterization of ozone production rate using satellite-derived columns of NO2 and HCHO along with modeled photolysis rates.  Overall, an excellent, important, manuscript, although it can be difficult to follow at times.

L206: At the beginning of the methods section, add a few sentences explaining how sections 3.1-3.4 fit together. 

L274-L296: How does the clustering described in section 3.3 fit into the rest of the paper?  Is it part of the coefficient determination or just an analysis tool. Please explain more clearly.

L400: How did you end up with 7 distinct classes after your clustering analysis?  Was it trial and error based on how the deviations of observations from the centroids of the 11 features  looked?

L524-543.  It appears that you adjust the TROPOMI NO2 and HCHO to remove biases with respect to MAX-DOAS and FTIR observations.  How important is this result to your bottom line coefficients for PO3 and wouldn’t these biases be regional and subject to change with new versions of TROPOMI data?

Minor Comments

L93 Break this paragraph into two with the last paragraph previewing what you are doing in this manuscript.

L127: Is there a version number for these recently reprocessed fields?

L199-200: I don’t understand the meanings of the colons within the parentheses.  Are these threshold values for the look up tables? If yes, why do the values jump around so much such as 100:50:600?

L268: Do you really mean equation 3 here?

Figure 2.  Perhaps italicize SZA, ambient temperature, and Pressure as they are dropped?

Figure 2: Why is the left column labeled “Input Candidates from Aircraft” when it uses model and satellite data?

L325: M2GMI (be sure to define somewhere)

L359-L364: Here or perhaps in section 3.2, Add some background on the role dilution factors play in box model calculations.

L391-397: The last 3 sentences of the Figure 3 caption contain information that is also in the main body of the article. Perhaps delete.  You probably should mention which field campaigns had the most observations and therefore played the largest role in determining the statistics.

Figure 4: I notice you use log(FNR) and log(FNP) here.  Could you explain the benefits of this transformation.

L449-451.  Did ambient T, H2O vapor, pressure and/or SZA add any additional insights? Preview the results here.

L467-469: Earlier you mention that SZA, pressure, and temperature were dropped.  Here you also include H2O vapor.

L621: Be sure to expand the Benelux acronym the first time it is introduced.

Figure 17.  You may want to change the order of the contributions so that the third listed contribution in the legend (jNo2) is also the third in the Figure (it is currently the second from the top).

L756-773: The financial support section lists numerous measurements some of which seem to have little relation to this project.  Would it be possible to tighten this section up by eliminating data sets that are only peripherally related to this study while adding more information on how particular measurements were important for this study.

Grammatical Comments:

L96: use degrees symbol.

Section 2.4.  …. But how do you convert the VCDS?

L197: To estimate photolysis rates of JNO2 and JO1d- To estimate the photolysis rates, JNO2 and JO1d), we

L216:  -->  (Tibshirani, 1996).  They consider a regression,

L292: These features include -->  These features are

L317: are based on converted the bias-corrected --> are derived by converting the bias-corrected

Figure 2: Typo.  Should be M2GMI Conversion Factor within the diamond.

L480: more photolysis rates --> higher photolysis rates

L487: by random dropping --> by randomly dropping

L513: predictor power --> predictive power

L596: making NO2 levels -->  meaning NO2 levels

L686: maps of within the PBL --> PBL maps