the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Jul 2026
Observations of radiocarbon (14C) in atmospheric CO2 confirm decline in U.S. emissions of fossil fuel CO2 (FFCO2) between 2010 and 2015
Abstract. All emissions pathways aimed at stabilizing global temperatures at the internationally agreed target of 1.5–2 °C above pre-industrial levels require steep cuts in global CO2 emissions. Reliable emissions tracking is therefore essential to monitoring progress towards related mitigation goals, especially for the world’s largest emitters. Here we make use of atmospheric measurements of Δ14CO2 and the dual-tracer Δ14CO2:CO2 assimilation and inversion system previously developed by our group to estimate annual and monthly CO2 emissions from fossil fuel use and cement production (FFCO2) for the U.S. for 2010 and 2015, the first two individual years for which large numbers of atmospheric Δ14CO2 measurements are available. Ensemble-mean national FFCO2 totals obtained from a 9-member suite of inverse results are larger than reported by the U.S. Environmental Protection Agency (EPA), but overlap at their respective 2σ ensemble-wide model spreads (inversions) and reported 95 % confidence intervals (EPA) in both years. In contrast, the inverse results agree with both annual totals and 16 of 24 derived monthly totals from the Vulcan 3.0 emissions data product with 1σ, ensemble wide. Central estimates of the change in U.S. FFCO2 emission between 2010 and 2015 range from -5.1 % (EPA), -5.8 % (Vulcan), and -7.8 % (this work), providing a first confirmation of an expected national FFCO2 decline based on atmospheric observations. The development of a reliable emissions tracking system based directly on atmospheric observations, as detailed here, may take on additional scientific and policy relevance given the recent interruption of EPA reporting.
- Preprint
(2905 KB) - Metadata XML
-
Supplement
(1832 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-2896', Anonymous Referee #1, 13 Aug 2026
-
RC2: 'Comment on egusphere-2026-2896', Anonymous Referee #2, 11 Sep 2026
The paper by Islam et al. provides estimates of fossil fuel emissions over the US for 2010 and 2015. The estimates are obtained by using an updated version of the dual tracer inversion method developed by Basu et al. 2016 and 2020. The estimates obtained from the inversion work are higher than the estimates from the US Environmental Protection Agency, and agree with those obtained by Vulcan. Furthermore the obtained top-down estimates for 2015 are significantly lower than 2010, hinting towards a decline.
This manuscript is a highly relevant study and a good demonstration of the application of atmospheric radiocarbon observations to verify fossil fuel emission estimates. I recommend publication after addressing two issues on which I'd like the authors perspective. Also, I highly recommend the authors to shorten the manuscript and make it more focused. I hope the authors find my line by line suggestions below relevant and helpful in order to achieve this.
Comments.The abstract highlights that the top-down estimate presented is larger than EPA and agrees with Vulcan. In their analysis of the US sub-regions, they show in table 3 that the western US region is significantly higher in the inversion results, which could fully account for the better agreement to the Vulcan totals. They also admit that this region is more uncertain, attributed, at least in part, to the amount of observations. Despite the smaller observational constraint over this region, the constraint is nearly independent as shown by table 5. Another potential cause suggested (maybe somewhat implicitly) is the existence of a bias in the Miller/CT2022 flux product used, as mentioned in line 593. However, the authors don't discuss the potential existence of a systematic bias of the inverse modelling system over this region. In fact, in line 509 the authors mention that the poor fit at THD might be due to transport model error, which might thus be larger in this region than it is in central and eastern US. For example, the orography differences could lead to regional systematic transport error which affects the west more than central and eastern US. I'd like to hear the authors perspective on this. Do the authors have reasons to believe that it would be small? whatever the answer, it should be discussed in text.
To aid in the discussion that addresses this, I propose an analysis on the fit statistics of the radiocarbon data on the same aggregate regions, west, central, and eastern US. For the western us region I suggest to leave out Alaska as the fossil emissions are smaller, and it is geographically remote to the other west coast sites which are more on top of the higher emission regions. I'm thinking of a barplot that shows per experiment the mean reasiduals of the fitted radiocarbon data. In addition, the mean residual per member could be plotted against the posterior flux results to assess a potential correlation between posterior fit statistics and the flux result. In this analysis it would also be helpful to see the posterior correlation between the disequilibrium biosphere flux of the different regions which currently is missing from table 5, but it is not stated explicitly that these are negligible. I agree that it would not demonstrate nor discard the potential existence of a systematic bias, but it is relevant to the interpretation. In addition this could also help with the discussion of the potential bias in the Miller/CT2022 product. I leave it up to the authors to decide, but I believe it will add to the quality of the paper.
The second comment relates to a claim in the title and abstract: "... confirm decline in US Emissions". Certainly the top down estimates of 2015 shows lower fossil emission estimates compared to 2010. Also, they agree with the shown EPA and Vulcan products in those years. However, figure 6 does not, in my opinion signal a decline. Hypothetically, when I leave out 2010 the FF emission appear rather stable to me. And so the current manuscript does not rule out the alternative hypothesis that fossil emissions are stable with 2010 being an exceptional year. The Vulcan product used has insufficient temporal coverage, but there are other validation datasets like Vulcan-NRT, GridFED, (see also my minor comment below), and EPA that can be included in a longer time-series that might actually demonstrate the decline. Alternatively, is there a reference to an analysis that shows fossil fuel emissions of the US to be in decline? Depending on the answer, adjust the text and title accordingly.
Finally, is there a reason why the authors did not use the freely available GridFED dataset (https://zenodo.org/records/17467681, Jones2021) in their comparison? - if there is a good reason, my suggestions related to the GridFED data below can be ignored.
Jones, M.W., Andrew, R.M., Peters, G.P. et al. Gridded fossil CO2 emissions and related O2 combustion consistent with national inventories 1959–2018. Sci Data 8, 2 (2021). https://doi.org/10.1038/s41597-020-00779-6
Line by line comments and suggestions:L35 First paragraph of the introduction reads a bit unfocused. Consider splitting up the first paragraph into clear focused paragraphs (i.e. lines 35--65): 1) the case for emission monitoring. 2) various methods (FFDAS/Bottom up/Carbon monitor etc). 3) the application of top-down and benefits. 4). central goal of the paper. Also Consider a few lines that explain the structure of the paper.
Line 76-95 contains information that belongs to, and is also reported on in the methods section; I suggest to leave it out of the introduction.
Line 101/Eq.1 I suggest leaving it out or moving this to the supplement. It is not used directly in the paper and serves no purpose. For people that work on top-down estimates using inverse modelling techniques this is known. For people that don't know these, it is insufficient to fully explain what inverse modelling does and what the role of the cost function is. It is beyond the scope of this paper to explain that, and handled in detail in other texts. I suggest instead to refer to a good overview to direct curious readers to derivation and application of the cost function in 4D-var system.
Line 140 Figure 1. Coastlines are missing for the panel a). Is this on purpose, a problem with my pc or a mistake?
Line 171 It is unclear if the priors taken from CT16 and CT19B are the posterior result of those inverse modelling estimates, or if those are the priors to these inversions? In case these are the posterior result of those experiments, they are estimates that used the same CO2 observations used in this work. This means that in 6 out of 9 members the solution space for the fossil fuel estimates are limited by prior NBE used. Is this justified?
Lines 176-181 It is currently implicit, but this essentially means that these fluxes are the same as Basu 2016/2020? Later it is claimed that the modelling of radiocarbon is a priori good because of the effort in Miller 2025a/b (line 490), where it is actually reported that the cosmogenic and terrestrial disequilibrium flux is scaled. Does that study use the same fluxes as Basu 2016/2020? I encourage the authors to leave no room for ambiguity.
Lines 203-214 Is this a result or discussion? It seems that this information is entirely repeated in the results section starting at line 248. Consider to remove it here, and condense the same information into a single discussion paragraph.
Line 215 I suggest to change "additional flux data sets" to something like "evaluation flux data sets", and potentially merge it with section 2.3.1.
Line 229-234 I think this does not serve a purpose in this paper. The OCO2 information is published in the aforementioned publications (line 226). Consider removing these lines.
Figures 2 and 3 can be combined and the EPA total could be added (and GridFED?); the resulting figure would demonstrate nearly all the results highlighted in the abstract: The higher emissions from top down compared to bottom up (EPA) and the better comparison to Vulcan as well as the noted decline. Also, consider in such an updated figure if showing the prior still serves a purpose. Adding GridFED would also be interesting to see, as it also has a seasonal cycle.
Line 346 The posterior covariance matrix relates to the prior but also depends on the sensitivity of the observations. It could be that, given sufficient observational constraints, the posterior length scale in both space and time is reduced compared to the a priori first guess. I agree that it can be used to regularize the solution in cases where these constraints are lacking, which likely is the case here. However, I'd be cautious stating that the chosen lengths scale would be in general limiting the solution. As a counterexample, what about in-situ co2 records and the temporal covariance in NBE? Or what-if more observations become available in later years?
Table 1: Given that you wish to compare to the CI of EPA and Vulcan, would it not be more clean to use the 2-sigma everywhere instead of reporting mixed 1-sigma and 2-sigma? I also suggest to make a choice between 1 and 2 sigma for the entire paper and stick with that. Finally, the authors could consider to moving this table to the supplement. The summary of these data is presented in Figures 2 and 3, and the table is somewhat redundant. The interested reader can decide to look up the numbers in the supplement if they want.
Section 3.4.1 and figures 7 and 8; Are these necessary? They confirm that the inversion system reported by Basu et al. 2016/2020 works as expected. Is the data for all ensemble 9 ensemble members or a single experiment? Also I fail to so see (but it could be me) the relevance of Figure 8, specifically the choice to show the gradient against some arbitrary altitude. What additional information does that provide compared to simply showing the residuals? I suggest moving this to the supplementary material. Instead I suggest replacing this by an analysis of the radiocarbon data as mentioned in my first comment. See also my comment to line 505 below.
Line 505 "This in turn suggests...." The study by Zhang 2026 conclude that large errors in vertical transport could exist of up to 4 ppm, and lead to aliasing in fluxes. I don't see how looking at the gradients would lead you to conclude that transport error leads to only modest aliasing. How do you dismiss the hypothesis that with the right forcing at the surface you could reproduce the gradients in the presence of transport error? Hypothetically, such an experiment can be done by deliberately changing values of certain parameters to simulate a systematic error in the transport model, and then redo the inversion. My hypothesis is that you will find flux-fields that match the presented gradients equally well, especially when assimilating the aircraft data itself. From such an experiment you may be able to quantify the effect of potential systematic transport errors on estimated flux-fields. Now, such an experiment is beyond the scope of the presented work. Moreover all inversion studies suffer from this problem. The authors refer to Basu 2016, where a 1) comparison with SF6 is made and 2) a detailed scheme of estimating R is discussed, for a discussion on the atmospheric transport (Line 109). If nothing is done in this work to improve transport, or quantify its error, I suggest to the authors that they simply stick to that reference which is good work, and admit that systematic transport errors pose an issue, but that quantifying them is beyond the scope of this research. If alternatively, the authors do have a good method for quantifying the potential aliasing using these gradients, I think it warrants more detail in text, and it could be highly valuable contribution so others can reproduce it.
Zhang, L., Davis, K. J., Schuh, A. E., & Jacobson, A. R. (2026). Impacts of transport uncertainty on CO2 over North America: Insights from assessments of two global models with aircraft and tower measurements. Journal of Geophysical Research: Atmospheres, 131, e2025JD044053. https://doi.org/10.1029/2025JD044053
Section 3.4.2 Would it be helpful to have a figure where the 9 posterior flux estimates are plotted against there a priori bio/fossil flux, colored by fossil/bio prior? This would immediately demonstrate a potential relation between prior and posterior flux.Line 592 Given that the Miller/CT2022 product cannot be objectively discarded, what is the relevance of the last row in Table 4? I suggest removing it as it adds noise to the paper. One could still mention that the estimated uncertainty is conservative because there is a hypothesized bias in the Miller/CT2022 product. Also the tabulated values in table 4 seem to be identical to table 1, and I wonder if it could be removed entirely.
Lines 620 - 630 it seems to me that this belongs in the methods section.
Lines 681 and Table 5 Why these regions? Are these the most relevant? are other correlations insignificant and small?
Citation: https://doi.org/10.5194/egusphere-2026-2896-RC2 -
RC3: 'Comment on egusphere-2026-2896', Anonymous Referee #3, 18 Sep 2026
This study of Islam et al. demonstrates a top-down framework for estimating fossil fuel CO₂ emissions using atmospheric radiocarbon (Δ¹⁴CO₂) observations, atmospheric transport modeling, and inverse methods. A key strength of the study is the relatively dense and well-distributed radiocarbon observation network across the United States, which provides sufficient spatial coverage to constrain national-scale fossil fuel emissions from atmospheric measurements. By applying the same methodology to two different years (2010 and 2015), the authors show that radiocarbon-based inversions can be used not only to estimate emissions independently of conventional inventories but also to quantify changes in emissions over time. The work highlights the potential of atmospheric radiocarbon observations as a verification tool for tracking long-term emission trends and evaluating reported reductions in fossil fuel CO₂ emissions.
The study builds strongly on the radiocarbon inversion for the United States previously published by Basu et al. (2020), while extending that work through methodological improvements, enhanced input datasets, and its application to two distinct years. Particular strengths of the study include (i) demonstrating the substantial value of the comparatively dense radiocarbon measurement network in the U.S., (ii) incorporating an ensemble of estimates based on multiple prior datasets, and (iii) providing a thorough uncertainty assessment by combining Monte Carlo-based estimates of posterior uncertainties with the spread across ensemble members.
The study is well within the scope of ACP, employs scientifically sound and well-established methods, and is presented clearly and logically. The results and discussion are comprehensive, although somewhat lengthy. Several datasets are analyzed repeatedly from different perspectives; however, these complementary analyses ultimately provide valuable insights into both the sensitivity of the results to the choice of prior datasets and the ability to quantify emissions separately across different subregions.
I recommend publication subject to minor changes as described in the following.
Main points:
- In my view, the authors are somewhat overconfident in the ability of the TM5-4DVAR system to quantify emissions with an accuracy of only a few percent, as suggested, for example, by the emphasis placed on the agreement with Vulcan (p. 9). While TM5 has been shown to reproduce large-scale transport reasonably well, it is subject, like all atmospheric transport models, to non-negligible systematic transport biases (e.g., Peters et al., 2004). Although the bias in interhemispheric mixing identified by Peters et al. (2004) was subsequently corrected, biases related to planetary boundary layer (PBL) mixing likely remain. Furthermore, Schuh et al. (2019) demonstrated substantial differences in the large-scale distribution of tracers between TM5 and GEOS-Chem, although translating these discrepancies directly into uncertainties in the FFCO₂ emission estimates presented here is not straightforward.
More generally, several studies have shown that even transport models driven by assimilated ECMWF meteorology can exhibit significant biases in PBL mixing and, consequently, in simulated CO₂ concentrations (e.g., Gerbig et al., 2008). Hu et al. (2014), for example, reported differences on the order of 20% in U.S. HFC-134a emission estimates obtained using two different Lagrangian transport models (HYSPLIT and STILT). Likewise, De Longeville et al. (2026) found systematic differences of approximately 10-15% between NAME- and FLEXPART-based inversions over Europe. Taken together, these studies provide little evidence that current atmospheric transport models can routinely constrain emissions with uncertainties below about 10%.
The ensemble approach adopted in this study, which explores different combinations of anthropogenic and biospheric prior fluxes, is a clear strength and provides valuable insight into uncertainties associated with prior assumptions. However, it does not address the issue of systematic transport biases, since all ensemble members rely on the same transport model. Consequently, the spread of the ensemble is unlikely to capture the full uncertainty of the inversion system.
In conclusion, I encourage the authors to be more cautious in their interpretation of the excellent agreement with Vulcan. While this agreement is certainly encouraging, it should not be taken as definitive evidence of emission estimates being accurate to within a few percent, as part of the agreement could result from a fortuitous compensation of errors between the inversion system and the inventory. - I am somewhat concerned about the choice of FFCO₂ and NBE prior datasets. Several of these datasets appear to originate from data assimilation systems (e.g., FFDAS v2 and CT2022 for FFCO₂, and CT2016 and CT2019B for NBE) that likely assimilated at least part of the same atmospheric CO₂ observations used in the present study. As a result, the prior estimates may not be fully independent of the observational constraints applied in the inversion. The authors should provide a stronger justification for this choice and discuss the potential implications for the interpretation of the results.
- There is little discussion of the inferred disequilibrium fluxes associated with heterotrophic respiration. How different are the posterior estimates from the priors? In addition, how sensitive are the results to the assumed isotopic (disequilibrium) signature of heterotrophic respiration? If this issue has been investigated in previous studies, it would be helpful to summarize the relevant findings and place the current results in that context.
- I found the discussion in the last section of page 9 regarding the influence of the SiB4 prior on the FFCO₂ flux estimates difficult to follow. While I agree that crosstalk between NBE and FFCO₂ flux estimates is expected, I would have anticipated that the stronger summertime NBE uptake in the SiB4 prior would lead to larger, rather than smaller, inferred anthropogenic emissions to compensate for the increased CO₂ drawdown. The authors should clarify the mechanisms underlying this result and explain why the opposite behavior is obtained.
- The influence of reduced heating demand resulting from the milder winter temperatures in 2015 relative to 2010 deserves a more prominent discussion. To what extent can the inferred reduction in CO₂ emissions be attributed to this meteorological effect, and how much remains attributable to structural changes or other mitigation measures? Providing a quantitative estimate of the relative contributions of these factors would strengthen the interpretation of the interannual differences.
Small points and typos:
- Page 2, line 57: You may add the study Gómez-Ortiz et al. (2025; https://doi.org/10.5194/acp-25-397-2025) even though it is only based on synthetic observations.
- P4, L117: Were the CO2 initial fields scaled in the same way?
- P5, L137: The number of observations (1615) is different from the one in Figure 1b (1582). Why?
- P5, L140: delete the second "(b)" in the figure capture.
- P6, L155: Change to "we specify three different"
- P6, L170: Change to "used to inform our prior"
- P7, L205: It would be good to know what a value of +118TgC/yr corresponds to in relative terms/percentage change.
- P8, L221: Is there only a problem of non-fossil emissions in Vulcan from biofuel combustion in transportation? What about biofuels/wood burning from residential heating?
- P9, L259: Change "increased" to "increase"
- P9, L265: It is not necessary to cite Gurney et al. over and over again when referring to Vulcan.
- P11, L314: Change to "we again see ensemble mean annual total"
- P13, L346: Change to "an assigned temporal"
- Figure 4: What are the light blue diamonds in the lower righthand part of the figure? I assume these are the priors, but this is not explained in the figure caption.
- Table 2: The units in these tables are not TgC/year but simply TgC or TgC/season
- P27, L713: The different observations sampling also contributed to the differences from the previous work.
- P27, L729: Why were there 15 tri-monthly seasonal totals and not just 2 years x 4 seasons per year = 8?
Citation: https://doi.org/10.5194/egusphere-2026-2896-RC3 - In my view, the authors are somewhat overconfident in the ability of the TM5-4DVAR system to quantify emissions with an accuracy of only a few percent, as suggested, for example, by the emphasis placed on the agreement with Vulcan (p. 9). While TM5 has been shown to reproduce large-scale transport reasonably well, it is subject, like all atmospheric transport models, to non-negligible systematic transport biases (e.g., Peters et al., 2004). Although the bias in interhemispheric mixing identified by Peters et al. (2004) was subsequently corrected, biases related to planetary boundary layer (PBL) mixing likely remain. Furthermore, Schuh et al. (2019) demonstrated substantial differences in the large-scale distribution of tracers between TM5 and GEOS-Chem, although translating these discrepancies directly into uncertainties in the FFCO₂ emission estimates presented here is not straightforward.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 797 | 132 | 48 | 977 | 97 | 43 | 52 |
- HTML: 797
- PDF: 132
- XML: 48
- Total: 977
- Supplement: 97
- BibTeX: 43
- EndNote: 52
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of „Observations of radiocarbon (14C) in atmospheric CO2 confirm decline in U.S. emissions of fossil fuel CO2 (FFCO2) between 2010 and 2015”
This manuscript presents results from a dual-tracer 14C-CO2 inversion for the years 2010 and 2015, when sufficient observational coverage was available across the U.S. target region. The dual-tracer inversion framework was developed by authors of this study in Basu et al. (2016, 2020) and has been refined for the present study primarily through the use of updated observations, meteorological fields, and prior emission estimates. The core of the dual-tracer inversion consists of 14CO2 observations, mainly from the dense NOAA station network in North America, which are used to separate fossil and biogenic CO2 fluxes in the U.S., thereby providing valuable additional information compared to a CO2-only inversion.
The inversion results indicate a decline in the U.S. fossil fuel CO2 emissions of more than 5% between 2010 and 2015, consistent with trends based on independent emission inventories. The derived ffCO2 fluxes are higher than those from (gridded) prior flux estimates but show impressive agreement with an adjusted, U.S.-specific emission inventory product (Vulcan 3.0), even at the monthly timescale.
The authors present their findings in a well-structured and comprehensive manner, supported by clear figures and discussions. A range of sensitivity analyses, based on varying prior fluxes and prior error correlations, along with model performance evaluations and a detailed analysis of posterior flux correlations, are conducted to assess the reliability and robustness of their ffCO2 emission estimates.
The study clearly demonstrates the potential of atmospheric 14CO2 observations for estimating national ffCO2 emissions and for detecting inter-annual trends. It underscores the critical need for continuous 14CO2 measurements to verify emission inventories, enhance confidence in emission reporting, and enable the monitoring of ffCO2 emission trends, which is relevant for assessing the effectiveness of climate mitigation strategies.
Overall, this is an excellent study, which I strongly recommend for publication after addressing a few minor issues outlined below.
Specific comments
L. 63-66: Given that the 14C-based ffCO2 emission estimates show better agreement with the Vulcan product than the EPA product in 2010 and 2015, it would be interesting to know which additional information is incorporated in Vulcan but not explicitly considered by the EPA.
L. 132-134: Consecutive hourly CO2 observations from a given site might exhibit correlated errors in the model-data mismatch. Do the authors account for these temporal correlations in the model-data mismatch error covariance matrix? Implementing such correlations might also prevent the large number of CO2 observations from exerting a too strong influence on the dual-tracer inversion results compared to the sparser 14CO2 data.
L. 135-137: What is the integration time of the Δ14CO2 samples? Were they collected only during the afternoon?
L. 151-153: What is the temporal resolution of the prior fluxes?
L. 171-173: If I understand correctly, two posterior NBE estimates from NOAA’s CarbonTracker system are used as prior constraints in the current study. I’m not sure whether the CarbonTracker system assimilates similar CO2 observations that are also used as observational constraint in the current study. If this is the case, it raises a concern about the potential double use of the same observational information, i.e., once in deriving the CarbonTracker-based NBE prior and again in the flux estimation of the current study. I’m therefore wondering about the independence of the CarbonTracker-based NBE priors, particularly in Northern America. Could the authors comment on this?
L. 203-205: For easier interpretation, could the authors also express the change relative to the earlier results (i.e., +118TgC/yr) in relative terms (%)?
Tab. 1: Interestingly, the spread among the national total ffCO2 posterior estimates is larger than the spread among the ffCO2 priors when the same NBE prior is used. In contrast, the spread in the posterior ffCO2 (2015-2010) trend estimates is smaller than the spread in the prior trend estimates. Could this indicate that the 14C-CO2 observations provide stronger constraints on interannual trends than on absolute annual total fluxes?
Fig. 4&5: Please note that the light blue diamonds within the gray shaded region of the figures represent the prior ffCO2 emissions.
L. 496-510: Did the authors also evaluate how well the prior CO2 values fit the observed vertical profiles from the NOAA aircraft profiling sites? When comparing the prior vertical gradients with aircraft observations, the simulation of the seasonally-varying vertical structure of the profiles remains unaffected by surface flux adjustments that may arise from the assimilation of the (same) aircraft data. Such an analysis would strengthen the statement made in L. 504-506.
Fig. 8: Please indicate in the figure caption that the simulations are based on posterior fluxes.
Tab. 4: Please specify the units of the values shown in the table.
L. 620-629: This section could be moved to the Methods section. Additionally, it would be helpful to provide a table summarizing the spatial and temporal correlation lengths assumed for each flux category in the default setup and the three sensitivity experiments (ST1-ST3).
L. 727-729: Why is April 2025 an exception? For me, this is not immediately clear from Fig. 3, where the 1σ uncertainty ranges of the prior and posterior means are clearly separated from each other.
L. 729-730: How do the authors obtain “15 tri-monthly seasonal totals”? In the main text and in Tab. 4, only 10 tri-monthly seasonal totals are presented, 5 for 2010 and 5 for 2015.
Supplement:
First paragraph: How large is the transport model error compared to the observational error for Δ14CO2? Is the transport model error time dependent?
Tab. S6: Where does the strong increase in the CT16 global NBE estimate from -1334 TgC/yr in 2010 to -266 TgC/yr in 2015 come from? Is this a potential artefact?
Tab. S6: Where do the isotopic values for the cosmogenic production come from? For example, Miller et al. (2025) report a value of 6282 PgC‰/yr for 2010 (in their Tab. 1), which was subsequently scaled by a factor of 0.8 to achieve global flux balance, resulting in roughly 5000 PgC‰/yr. However, in Tab. S6 of the current study, a value of only 466 PgC‰/yr is assumed for 2010. How do these values align?
Technical corrections
L. 170: “to inform our prior fluxes, …”
L. 259: “increased” -> “increase”
L. 271: “FFCCO2” -> “FFCO2”
L. 317: “±26 TgC/y” -> “±26 TgC/yr”
L. 346: delete “a”
L. 583: “a prior” -> “a priori”
L. 646: delete one “(“
References:
Basu, S., Miller, J. B., and Lehman, S.: Separation of biospheric and fossil fuel fluxes of CO2 by atmospheric inversion of CO2 and 14CO2 measurements: Observation System Simulations, Atmos. Chem. Phys., 16, 5665-5683, 10.5194/acp-16-5665-2016, 2016.
Basu, S., Lehman, S. J., Miller, J. B., Andrews, A. E., Sweeney, C., Gurney, K. R., Xu, X., Southon, J., and Tans, P. P.: Estimating US fossil fuel CO2 emissions from measurements of 14C in atmospheric CO2, Proceedings of the National Academy of Sciences, 117, 13300-13307, 10.1073/pnas.1919032117, 2020.
Miller, J. B., Lehman, S. J., and Lindsay, C. M.: Numerical Representation of Contemporary Atmospheric Δ14CO2: 1. Time-Varying Global Fluxes and Atmospheric Mass Balance, Global Biogeochemical Cycles, 39, e2025GB008522, https://doi.org/10.1029/2025GB008522, 2025.