Review of Guo et al., Heterogeneity and Chemical Reactivity of the Remote Troposphere defined by

Aircraft Measurements

The manuscript presents the development of a gap-filled database of observations from the Atmospheric Tomography (ATom) aircraft campaign, designed to provide unbiased measurements of the chemical composition of air over the Atlantic and Pacific ocean basins over a large vertical fraction of the troposphere. The paper compares the observed concentrations and 24-hour averaged photochemical fluxes (ozone production, ozone destruction, methane destruction) calculated from six different global chemistry models constrained by the observations with the same quantities sampled from freely-running simulations of these same models. Of particular interest, the comparison of observations and models is done in a statistical sense and aims to address the issue of whether global chemical models run at the resolutions currently used (1 or 2 degrees in the horizontal) are sufficient to resolve the distribution of photochemical reactivity seen in the much higher spatial resolution aircraft observations. The authors find that models should be able to reproduce the distribution of photochemical reactivities, but also find significant biases in the concentration of NOx that results in biases of the distribution of, in particular, ozone production.

The ATom observations are a fantastic addition to the set of measurements of the chemical composition of the troposphere we have and the approach of statistically comparing observations and models allows us to advance past the facile comparisons of long-term means or requiring models put the right plume in the right place at the right time. While the approach and the results have tremendous promise, the organization of the material makes it very difficult for a first-time reader to make sense of it. For example there are initial references to RDS_R0, RDS_R1 and RDS_R2 (Line 153, Table 1) without any supporting explanation, forcing the reader to search through the Supplementary Information or be patient to find some discussion of these differences between these data streams in the Results section. There is almost no discussion of how the RDS is calculated except for a generic reference to the Supplementary Information. Taking some of the text from Section S.2 (starting at line 428 of the Supplementary Information) would help to improve the ability of the reader to understand the manuscript. And while many of the models that are used to calculated the RDS are well known, there is the use of F0AM which is a box model evidently, but with no other information, and which has been run using ‘MDS’ (Table 1), which I assume is the Model Data Stream?  Other details that would help in the interpretation of the results, such as whether the dates for the driving meteorology for the Chemical Transport Models (CTMs) and for nudging the Chemistry Climate Models (CCMs) match the ATom measurement dates, can only be inferred. Many of the minor comments are directly related to the problems of organization of the presentation and, while individually these are rather nit-picky concerns, the cumulative effect on the reader is disorienting and it takes considerable effort and searching to understand the material being presented.

My other significant concern is with one of the fundamental results of the paper – that the spatial variability in reactivity calculated from the original ATom data should be resolvable by our current global CTMs and CCMs. The results are discussed in lines 345 – 353 and shown in Figure S8. Perhaps it is a problem of my understanding as the approach to calculating the results for Figure S8 is not explicitly stated: I assume you took random points from the P-O3 frequency distribution and then averaged the 10s data points that were adjacent in space and time along the ATom flight path? If the length scales inherent in the data were of the order 100 km, and thus resolvable by models, then I would think the frequency distribution would not change very much as three or five points adjacent in space are averaged. But Figure S8 shows a rapid change of the distribution towards a Gaussian centered on the mean of the original 10s data. So I am not able to understand how the results shown in Figure S8 support the idea that models should be able to resolve the spatial scales found in the ATom-derived reactivity. The results from averaging 8 adjacent data points (~16 km) shows almost no occurrence of P-O3 greater than 4 ppb/day.

Minor Comments:

Lines 144 – 156: This section has a discussion of the need for gap filling and introduces Reactivity Data Stream RDS_R0. There is also reference to Table 1 where RDS_R0, RDS_R1 and RDS_R2 appear. But the reader must dig into the Supplementary Information for any idea of what RDS_R0, etc. refer to. There is not even mention that the nomenclature refers to different approaches to deal with data gap filling. It is in the Results section, starting at line 167, that there is some discussion of R0, R1 and R2. The history of the different MDS versions and how they led to different RDS versions needs to be coherently introduced. 

Lines 151 – 162: This section describes the RDS, but there are no details on how the RDS is calculated, forcing the reader to go to the Supplementary Information or back to Prather et al. (2017). The text of this manuscript should include sufficient details to allow the reader to make sense of the article as they read through it so I would urge the authors to include some minimal outline of how the RDS is calculated by the different models using the MDS as input. 

Lines 164 – 165: The reader is referred to Methods for information on how the MDS was constructed. There does not appear to be a Methods section in the body of the article. Do the authors mean to refer to the Supplementary Information?  For that matter, there is also reference to a Methods section at lines 113 and 116.

Lines 186 – 187: ‘We include the statistics from UCI using alternate years (1997 and 2015

versus the standard 2016) to show the effect of different cloud fields..’ underlines the lack of details in the manuscript about how RDS is calculated. It is not mentioned anywhere that I can find what years were used for the RDS calculations.

Lines 228 – 231: The authors find that the model calculations of RDS show quite similar distributions of reactivities when constrained by the MDS (Figure 1). How does meteorological variability fit into this comparison. There is not much information on how the RDS was calculated – time and space matched to the ATOM flights using CTMs and nudged GCMs, sampling a number of different years or just a single year – so it is difficult to judge. 

Lines 293 – 294: ‘The complex patterns of the 3Rs seen in Figure 2 cannot be matched directly with CCMs’. From Table 1, at least two of the CCMs were nudged to reanalysis. Would that not provide a similar level of fidelity for transport and airmass history as the CTMs? Later, at lines 318-319, there is the mention ‘this could be tested with CTMs using 2016 meteorology and wildfires.’ so it seems even the CTMs were not run with meteorology specific to the ATom campaigns? This is all well and good, but another example of the way in which a very ‘thin’ description of the setup makes it very difficult to interpret the results.

Line 303: A duplicate ‘that’ in ‘indicate that that P-O3’

Lines 328 – 344: This paragraph discusses Figure S7 that is found in the Supplementary Information. I would suggest moving Figure S7 to the main body of the article if you are going to discuss it at any length.

Lines 345 – 353: As for Figure S7, I would suggest Figure S8 move from the Supplementary Information to the main body of the article.

Lines 372 – 375: On the disagreement for HOOH (‘ If anything, the models tend to have too much HOOH: ATom shows systematically large occurrences of low HOOH (50-200 ppt, especially Central Pacific) indicating, perhaps, that convective or cloud scavenging of HOOH is more effective than is modeled.’) I agree the scavenging could certainly be the source of the problem. And, while it is equally speculative, I can’t resist pointing out that an overestimate of HOOH photochemical production would agree with the low bias for NOx found in the models. Is there any correlation between the regions where HOOH is overestimated and NOx is underestimated?

Line 615: I was not able to find any captions for the three tables.

Line 615: In Table 1,two of the three CCMs explicitly mention nudging for meteorology but NCAR (CAM4-Chem) just says ‘MERRA’. Was it also nudged to MERRA?

Supplementary Information

Lines 109 – 110: It takes digging into Table S2 to deduce that the NOx (PSS) calculated for MDS_R0 seems to refer to the calculation of the NO2 concentration from measured NO and assuming PSS. Starting at line 194 we learn a little bit more about the problem with NO2, but it is still not quite clear how NOx for the the final MDS_R2 was calculated. Were these data points dropped? The description of this problem, in particular, should be a little friendlier to the reader who is coming to this data for the first time.

Line 258: What is CO_N in ‘Create a continuous CO_N record.’?

This manuscript presents (1) several gap-filling methods for the ATOM dataset, creating a model data stream (MDS) (2) methods for calculating reactivity data streams (RDS) using the MDS, and RDS for six models, (3) a comparison of modeled and measured reactivity, with a specific emphasis on the ability of coarse resolution models to capture observed spatial heterogeneity. The authors conclude that models are capable of reproducing the statistical distribution of measured reactivities.

The work closely follows the analysis in Prather et al. (2018), and adds a powerful observational dataset. The MDS and RDS datasets are valuable to the wider atmospheric chemistry community, and the approach has a solid foundation. Further clarification of methods and a more careful quantitative analysis would greatly improve the manuscript. Comments below refer both to content and clarity.

Major:

• It would be helpful if the key reactivities were defined early in the manuscript rather than in the supplement.
• The F0AM model can be configured with a number of chemical mechanisms (i.e. F0AM itself does not have reactions, but relies on the MCM, GEOS-Chem, Carbon-Bond, etc). More specification on the F0AM setup is needed. For example, in lines 215-216, the authors refer to the “F0AM protocol for NOx”—where does this protocol come from? It seems the box model could be set up such that NOx can photochemically evolve. The rational for discrepancy in model procedures need further explanation.
• It is unclear why the RDS_R0 is used when it contains known errors, that were later fixed (lines 171-172). It seems using the most accurate RDS is possible (lines 180-181), and it would yield the most useful paper.
• One of the supporting pieces of evidence that that models can capture spatial heterogeneity is the descent given in the Supplement Figure 2. The analysis of this is purely visual. It looks as if some of the reactivities may vary by up to 50% in a given 500 m box. What is the variability? What would be considered an “acceptable” level of heterogeneity in a box?
• Line 287 says “the spatial scales of variability are within the capability of modern global models”, but directly after, line 293 says “the complex patterns of the 3Rs seen in Figure 2 cannot be matched directly with CCMs”. It seems these two statements are incompatible. Can you clarify?
• Paragraph starting at line 345: It is unclear to me how this analysis and Firgure S8 supports the conclusion that “the ability to nearly match Atom-statistics is […] significant” (also, what is meant by a significant ability?). Perhaps a dummy argument would help.
• Paragraph starting at line 354: The authors lead the reader to assume the models are missing a lighting NOx source. This would be a major conclusion that needs to be placed in the context of other literature. But also, I am wondering about the discrepancy in NOx between MDS versions discussed earlier in the manuscript. Has that impacted this analysis? The same comment applies to the following paragraph. Clarifying notation and using the most accurate datasets would help readers.
• The conclusions sections presents two new figures (figure 5, figure s6) and two new “quick look” interpretations. These brief analysis are cursory, and not conclusions of the paper.

Minor:

• Line 92: “most models agree in the CH4 and O3 chemical budgets”: does this mean “terms in budgets” or budgets themselves?
• Line 170: Comments like “Three central models showed excellent agreement” are vague. What agreed? How do you quantify that agreement?
• Line 183: “In our analysis, the ATom 10s parcel s are weighted to achieve uniform sampling”: What does this mean? Is there some post-processing weighting of the observations?
• Line 197: “Key photolysis rates are similar across all model except GISS, and because of this and other inexplicable results…” Should we assume that GISS model is fundamentally different than the others, or that there was some unexplained error in the model setup? 

References:

Prather, M. J., Flynn, C. M., Zhu, X., Steenrod, S. D., Strode, S. A., Fiore, A. M., Correa, G., Murray, L. T., and Lamarque, J. F.: How well can global chemistry models calculate the reactivity of short-lived greenhouse gases in the remote troposphere, knowing the chemical composition, Atmos. Meas. Tech., 11, 2653-2668, 10.5194/amt-11-2653-2018, 2018.