Summary:

Kim et al. present method for source apportionment of urban CO2 sources in the Greater Los Angeles (LA) area based on atmospheric observations of CO, CO2, 13CO2 and 14CO2. They use the difference of each species compared to background observations at MWO  to infer contributions from biospheric CO2 and CO2 from fossil fuel burning further separated into CO2 from gasoline and natural gas combustion.

The authors find significant contributions from natural sources in the LA area as well as seasonal changes in the contribution from natural gas and gasoline combustion. Lastly, the study briefly assesses the uncertainty of different source apportionment methods.

Overall, the paper is straightforward, well-written, easy to follow and clear in its methods, results and discussion. The method is an advancement of previously published source apportionment methods and will contribute to the continuously growing field of research focussing on greenhouse gas emissions form urban areas, which will surely be of interest to the readers of ACP. Before publication, however, the issue of the petroleum source signature should be addressed as well as some other minor and technical comments.

General comments:

The choice of d13Cpet as 25.5+/0.5permil is unclear. This is not the exact number suggested for CO2 from car exhaust by Newman et al. 2008 (see comment L 183). Furthermore, it would be prudent to validate the value of d13Cpet given in Newman et al. as the sources of crude oil being refined in California has changed significantly over the last 20 years. Alternatively, at least an assessment of the impact a different source signature would have on the results should be added.

Specific/technical comments:

L40: bottom-up emission inventories/models do not solely rely on consumption data as suggested here but can also use activity data to estimate emissions. For on-road vehicle emissions, vehicle kilometers/miles travelled (VKT/VMT) is a common proxy activity used.

L46: there is important work preceding Heimburger et al. e.g. : Mays et al. 2009 https://doi.org/10.1021/es901326b; Cambaliza et al. 2014, https://doi.org/10.5194/acp-14-9029-2014

L48: Preceding work by Breon et al. 2015 https://doi.org/10.5194/acp-15-1707-2015; is not mentioned

L57: How can Hardiman et al. 2017 be cited as evidence that recent improvements in biosphere models like Gourdji et al. 2022 are insufficient, given that Hardiman et al. was published 5 years before the improvements by Gourdji et al. were?

L183: Why was a value of -25.5+/-0.5 permil chosen for d¹³C_pet?

Table A1 in Newman et al. 2008 lists 2002-2003 petroleum composite as -25.8+/-0.5permil, but automobile exhaust was reported at -26.0+/-0.2permil. Furthermore, this signature strongly depends on the source of the crude oil (local less depleted vs important more depleted crude). The local California field production has fallen from 700k bbl/day in 2002-2003 to ca. 300k bbl/day 2022-2023 according to https://www.eia.gov/, thus the gasoline d13C has likely changes in the last 20 years.

(see also attached figure from https://www.energy.ca.gov/)

Also, Bush et al. 2007 reported for gasoline in Utah ca. -26.7permil to -28.8permil (table 2) and ca. -28.4permil for exhaust CO2 when combusted in a car engine. https://doi.org/10.1016/j.apgeochem.2006.11.001

Which is similar to work a decade later outside the US, e.g. Pugliese et al. 2017 reported ca. -27.5permil to -28.5permil for an extensive sampling of gasoline and other petroleum products in Eastern Canada https://doi.org/10.1016/j.apgeochem.2016.11.003

L231: please add % after every number that is expressed in percent. (Like done in line 243)

L250: ppb ppm-1 is not exactly the unit here as you comparing moles of CO per moles of air with moles of CO2 with moles of air. Suggestion to either change unit to ppbCO ppmCO2^-1, explain your shorthand briefly or change the title of the section to “CO:CO2 ratio (R) values of…” to avoid confusion for the reader.

L465: Figure 1. The labels for GRA, USC and Ful have little contrast, suggestion to change color.

L500: The labels on the ordinate overlap with the units and the abscissa lacks the label (date in xx/xx)

This paper describes the partitioning of CO₂ enhancements over Los Angeles (relative to incoming air) into biogenic, petroleum and natural gas components, using a combination of three different tracers – CO, ∂¹³CO₂ and ∆¹⁴CO₂.  This is a very nice study presenting a clever new way to partition emissions, and is entirely suitable for publication in Atmospheric Chemistry and Physics.  I recommend some fairly minor changes prior to publication.

Specific comments:

Please check for subscripts/superscripts throughout, and there are a few grammatical errors that should be readily fixable with a careful reading or two.

Lines 19-21.  It would be helpful to give the typical magnitude of the CO₂xs signal to put the uncertainty in CO₂ff into context.  From Figure 4, it looks like the typical magnitude is ~20 ppm so 3.2-4.8 ppm is a relatively large uncertainty.

Lines 38-39.  The authors may want to refer to efforts such as the IG³IS Urban Guidelines here, and perhaps reference other papers that have demonstrated the need for monitoring systems.

Turnbull JC, DeCola PL, Mueller K, Vogel F, Agusti-Panareda A, Ahn D, Baidar S, Bovensmann H, Brewer A, Brunner D et al. 2022. IG³IS Urban Greenhouse Gas Emission Observation and Monitoring Good Research Practice Guidelines - WMO GAW IG³IS Report 275, 2021. Geneva: World Meteorological Organisation.

Lines 39-59.  The references given in this section are almost exclusively examples from the US.  Suggest adding some examples from other regions of the world.

Lines 53-55.  Can you clarify this statement – I think I understand that these previous studies have assumed that the biogenic emissions are “known” and the inversion has therefore solved only for fossil fuel CO₂?

Line 63.  Reference Stuiver and Polach 1977.

Stuiver M, Polach HA. 1977. Discussion: Reporting of ¹⁴C data. Radiocarbon. 19(3):355-363.

Lines 74-76.  There are several examples of using R_ff for aircraft campaigns, where time variability in R_ff is not a concern, such as:

Graven HD, Stephens BB, Guilderson TP, Campos TL, Schimel DS, Campbell JE, Keeling RF. 2009. Vertical profiles of biospheric and fossil fuel-derived CO2and fossil fuel CO₂ : CO ratios from airborne measurements of Δ¹⁴C, CO₂ and CO above Colorado, USA. Tellus B. 61(3):536-546.

Turnbull JC, Karion A, Fischer ML, Faloona I, Guilderson T, Lehman SJ, Miller BR, Miller JB, Montzka S, Sherwood T et al. 2011. Assessment of fossil fuel carbon dioxide and other anthropogenic trace gas emissions from airborne measurements over Sacramento, California in spring 2009. Atmospheric Chemistry and Physics. 11(2):705-721.

Lines 79-80.  There is also a potential CO source from oxidation of VOCs, particularly in summer.  See for example:

Vimont IJ, Turnbull JC, Petrenko VV, Place PF, Sweeney C, Miles N, Richardson S, Vaughn BH, White JWC. 2019. An improved estimate for the ¹³C and ¹⁸O signatures of carbon monoxide produced from atmospheric oxidation of volatile organic compounds. Atmospheric Chemistry and Physics. 19(13):8547-8562.

Lines 81-82.  Dividing CO₂ff into high and low CO sources has also been done for urban inversions, and it has worked quite well.  See:

Lauvaux T, Gurney KR, Miles NL, Davis KJ, Richardson SJ, Deng A, Nathan BJ, Oda T, Wang J, Hutyra LR et al. 2020. Policy-Relevant Assessment of Urban CO₂ Emissions. Environmental Science & Technology. 54(16):10237–10245.

Line 155.  Reference Craig 1957 or other suitable paper.

Craig H. 1957. Isotopic standards for carbon and oxygen and correction factors for mass-spectrometric analysis of carbon dioxide. Geochimica Et Cosmochimica Acta. 12:133-149.

Lines 155-158.  I would expect that by calculating ∂_src on a sample-by-sample basis, samples for which CO₂xs is small will have very large uncertainties.  Is that the case, and how do you deal with that?

Line 176.  Please add a sentence summarizing the result of the test of using a different ∂bio value.  Currently there is only a figure in the supplement, but no explanation in the main text.

Line 182.  Does the ∂pet value include all petroleum sources?  Are there differences in ∂¹³C for diesel, gasoline and other products?  What is the ∂¹³C of biofuel additives?  Are they large enough to matter?  Have they changed through time? 

Line 221-223.  How exactly are biofuels (particularly ethanol in gasoline) accounted for in this multi-tracer analysis?  This could be complicated, because ¹⁴C will see biofuels as a biogenic source, but CO will see biofuels as a petroleum source since they will have a high R value.  Please explain more carefully in the text.

Lines 241-249.  It would be useful to include a direct comparison of the relative contributions of petroleum and natural gas between the observations and Hestia.  Do they agree?

Line 256.  Table 2, not Table 1?

Lines 268-270 and Table 1.  The summed R_pet and R_ng look reasonable, but the R values for each sector in Table 1 don’t seem to make sense and can’t add up to the summed value given.  Onroad, for example, shows R of 31, but by my calculation should be 14.

Also, did you consider using the CARB CO inventory?  In general, CARB does a much better job for CO than the NEI, but it may be that the NEI has simply adopted the CARB CO values for California.

Line 273.  I think the 28 % and 60% are the other way round.

Lines 285 – 302.  I don’t quite understand how this analysis is done.  Is it that you determine R for each week separately, and apply the R value for a given week.  Then calculate R for each 2 week period, and apply to all the measurements in that 2 week period.  Then repeat for longer and longer periods?  How is the uncertainty calculated?  From what is currently presented in the paper, I would interpret Figure 7 differently:

Figure 7 shows low uncertainty in CO₂ff when a single week is used, and then a much larger uncertainty for all longer periods, with a gradual improvement in uncertainty (for the CO + ¹³C method) as more weeks are averaged. 

But when R is calculated for a single week, there are only one or two data points to constrain the R value, I think.  So I wonder if the low uncertainty calculated when only a single week is used is artificial due to having little data to test against.  The noise as R increases to 2, 3, 4, 5 weeks is most likely an artifact of having small datasets, and smoothing of the uncertainty as the temporal resolution of R increases makes sense. 

Thus my reading Figure 7 is that determining R over a longer period is better than using short averaging periods, probably simply due to the small number of flask measurements available for short averaging periods.  Further, I don’t see any convincing evidence that R should change over periods of weeks or months, although R almost certainly has changed over years/decades as air quality controls have improved.  Given the large uncertainties in CO₂ff derived from ¹⁴C and the small number of flask measurements, variability in R over the short term is more likely explained by that uncertainty than by real variability in the R value.