the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Sep 2025
Uncertainty and retrieval sensitivity in TROPOMI-based methane inversions over the North Slope of Alaska
Abstract. The Arctic is experiencing unprecedented environmental changes with rapidly rising temperatures. Emissions of methane (CH4) – a potent greenhouse gas – may be increasing from the region, making accurate monitoring essential. The TROPOspheric Monitoring Instrument (TROPOMI) instrument offers high spatial and temporal coverage of CH4 column mole fractions. However, its data in the Arctic has historically exhibited seasonal and latitudinal biases and low-quality retrievals. A major challenge is the lack of ground-based validation data in high-latitude regions, which are used to improve satellite retrievals. This study evaluates inverse modelling to estimate CH4 emissions using TROPOMI measurements over the North Slope of Alaska. Using two retrieval products – the operational SRON product and the scientific WFMD product from the University of Bremen – we assess the alignment of derived emissions with surface measurement-derived inversions over 2018–2020 and test their robustness through sensitivity analyses. Our results show that tundra emissions from SRON inversions align more closely with surface measurement-derived emissions than WFMD inversions. Both TROPOMI-product derived emissions have anomalously low emissions in August 2018 compared to surface measurement-derived emissions, likely due to low data density resulting from high cloud cover. TROPOMI inversions provided stronger constraints on fugitive anthropogenic emissions compared to surface inversions. However, each retrieval produced different emission estimates, highlighting retrieval-dependent differences. Sensitivity tests revealed a strong prior dependence in both retrievals, raising concerns about robustness in northern high latitudes. This study highlights the importance of using multiple retrievals and rigorous sensitivity testing in high-latitude satellite inversions.
- Preprint
(2471 KB) - Metadata XML
-
Supplement
(7884 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-4230', Anonymous Referee #1, 29 Oct 2025
-
AC1: 'Reply on RC1', Rebecca Ward, 16 Dec 2025
Below is our response to reviewer 1. Our comments are in italics.
1) General comments.
Methane is second most important greenhouse gas in Earth's atmosphere, and the Arctic region is a potential large source of CH4 emission. Emission detection and monitoring is therefore an interesting and relevant research question. The authors compare emission estimates based on surface-only and TROPOMI based inversions using two different retrieval products. A sensitivity study is conducted to asses the robustness of the results. Since the article adresses a relevant sicientific topic from an inversion modelling perspective, it should be published after consideration of the specific comments given below.
Thank you for your positive review of our work, which has helped to improve the manuscript. We respond to your specific comments below.
2) Specific comments
*) page 4, line 99: the inversion method used in this study is the same as in Ward (2024). A few lines about the general setup of this method would make the rest of the section easier to follow for those readers who are not familiar with this method and don't have the time to read the referenced paper (e.g. what is the "tuning phase" (line 115) or "MCMC sample" (line 119), or how is the uncertainty interval (line 118) calculated?).We have substantially revised this section to provide a self-contained description of the HBMCMC inversion method, so that readers do not need to consult Ward et al. (2024). This includes the addition of an equation for the hierarchical Bayesian framework, a clearer description of the tuning phase, and an expanded explanation of the MCMC sampling procedure and uncertainty estimation. All changes are documented in the tracked changes version of the revised manuscript.
*) page 5, line 108: in some models negative emissions are interpreted as uptake instead. Why is that not considered here?We use truncated normal prior PDFs for both emissions and boundary conditions to prevent the sampler from exploring non-physical or extremely large negative flux values. This choice reflects that CH4 net uptake is expected to be very small relative to emissions at the spatial and temporal scales of our inversion. In practice, uptake can still be expressed as a reduction from the prior to the posterior (i.e., posterior scaling factors below one), even though absolute negative fluxes are not allowed. We agree that prohibiting grid-cell–level negative fluxes is a limitation, and we have now clarified this point in the manuscript: “To avoid sampling large, non-physical negative emissions, we use a truncated normal prior PDF for both the emissions and boundary conditions. This constraint means that net uptake cannot be represented at the grid-cell level, although reductions from the prior can still capture relative decreases in emissions.”
*) page 6, lines 156-158: I interpret regridding as calculating the average of all observations falling into the gridcell of the mentioned dimensions. This already reduces the data density of the observations, so why is additional reduction step necessary? Would using larger regridded "observations" without the additional reduction step have the same effect as the current approach?
Regridding to the NAME resolution (0.234° lat × 0.352° lon) reduces data density but unfortunately still leaves too many observations to run NAME within a feasible time frame. The additional reduction step was necessary to reduce the number of backward simulations further, whilst still allowing us to keep the most important observations over the North Slope of Alaska. We note also that Thompson et al. (2025) faced the same issue using a different Lagrangian particle dispersion model, FLEXPART, with TROPOMI and combined spatial averaging with an error-based thinning routine to manage the computational expense.
*) page 9 line 203-205: the XCH4^{model}_{pert} is apparently valid from surface to maxlev (right-hand side of supplement eq 4 = left-hand side of supplement eq. 7). That should be made clear in the text as well, since now it's just "we have equation 2 and subtract two other terms". So these lines could be clarified by first stating why you would need XCH4^{model}_{pert}, and then how you calculate it.This is a good point and an explanation of explicitly why we need XCH4^{model}_{pert} will help readers. We now include at the beginning of this paragraph: “As the inversion requires modelled mole fractions that are only dependent on NAME sensitivities, we construct a perturbed modelled column mole fraction, XCH_{4,pert}^{model} |{t}. This represents the averaging-kernel-weighted NAME contribution over the column, from the surface to the maximum level. We obtain XCH_{4,pert}^{model} |_{t} by subtracting all the retrieval-provided prior contributions, as follows, CH^{prior}_{4,i}, A_{i} and p_{i} are provided from the retrieval product, so the second term of Eq. \ref{eq:modelled_ch4_conv} can be subtracted from XCH^{model}_{4} |_{t}. Additionally, because NAME is only run up to 20km altitude, we assume the prior profile above this level and we can also subtract this known contribution from XCH^{model}_{4} |_{t}.”
*) page 9, equation 4: if the first summation is from "1 till maxlev", shouldn't the second one be from "maxlev+1 till n"? In other words, which of the two summations includes maxlev? Related to that is equation 3 and the rest of the derivation in the Supplement.Great point. We have updated the derivation to make it clear when the summations include maxlev and when they start from maxlev+1.
*) page 9, line 220: please clarify the CAMS dataset that you're using and why you use an outdated version (based on the "data availability" section, you could have downloaded one in 2023). The problem with a "CAMS dataset" is that there are multiple datasets providing methane estimates, see
https://ads.atmosphere.copernicus.eu/datasets?q=methane&limit=30
Neither of the datasets listed here is from 2019. The only dataset that I found using a "v19r1" or similar versioning scheme is called "CAMS global inversion-optimised greenhouse gas fluxes and concentrations" for which the "Evaluation and quality assurance (EQA) reports" on the documentation tab point to a report by Segers, not to Inness.The one you found is indeed the version and type of CAMS product we used. We have now been more specific about the version and type of dataset in the manuscript text, went back to find exactly when this data was accessed and added a more appropriate citation for the dataset.
“A priori vertical boundary conditions are derived from the CAMS global inversion-optimised greenhouse gas fluxes and concentrations v19r1 product (Segers and Houweling, 2020).”
“Segers, A. and Houweling, S.: CAMS global inversion-optimised greenhouse gas fluxes and concentrations, version v19r1, Atmosphere Data Store (ADS), https://doi.org/10.24381/ed2851d2, accessed: 2020-12-28, 2020.”
The BRW inversions began in 2020, and we use the same boundaries for consistency across the two papers (this one and Ward et al., 2024). That said, it is of course pertinent to use up-to-date versions, and future inversions will use more recent boundary-condition datasets. In Section 5 we already state: “...the boundary conditions are optimised each month throughout the inversion, and the primary aim of this study is a comparison of satellite and surface observations in the same inversion framework, rather than to quantify absolute emissions. Future inversions will use more up-to-date versions of CAMS.” Rerunning the inversions with a more recent CAMS inversion product is unfortunately not within the scope of this manuscript, owing to the lead author’s move to a new position.
*) page 10, line 229: why use EDGAR v6? As you mention in section 4.1(line 492-494), emission totals over Alaska may vary significantly according to the version of the database being used.A fair point and more up to date versions of EDGAR are now used in current inversions. If the aim were to define an inventory estimate for Alaska, for example, it would be essential to use a more recent EDGAR version. Given that this work is framed as a sensitivity study, we consider the use of EDGAR v6 acceptable. Finally, as with the previous comment, rerunning the inversions with a more recent EDGAR version is not within the scope of this manuscript, owing to the lead author starting a new position.
3) Technical corrections
*) page 1, line 80: Since the acronym HBMCMC is used here, I would capitalise the word "hierarchical" as well.Done
*) page 5, line 125: the 78W should be 78N?Corrected
*) page 6, line 156: please clarify if the regridding resolution is latitude x longitude or the other way around? The same holds for the other spatial resolutions mentioned hereafter.Latitude and longitude are added here and for all other spatial resolutions given.
*) Supplement page 1, eq. 6: the second summation term should have n instead of 1 as it's upper levelWell spotted, thank you.
*) page 19, line 387-388: there is no figure 9A or 9B.References to these figures are removed.
Citation: https://doi.org/10.5194/egusphere-2025-4230-AC1
-
AC1: 'Reply on RC1', Rebecca Ward, 16 Dec 2025
-
RC2: 'Comment on egusphere-2025-4230', Nicole Montenegro, 14 Nov 2025
Overall, the study appears to be well justified. The research gap is clearly articulated, as is the importance of investigating this topic and the substantial challenges associated with it. The use of recent observations positions the study as a meaningful and timely contribution, particularly in regions known to contain significant gaps in observational coverage and monitoring infrastructure, and which represent an important source of methane emissions.
I think it is a great contribution that the study examines uncertainty through the sensitivity of inversions to various configurations. However, I believe that these results could have been given greater prominence, especially considering that one of the stated objectives of the manuscript is precisely the quantification of uncertainty. The analysis places more emphasis on RMSE reduction than on RMSE spread, which is the quantity that more directly reflects uncertainty. I expected the manuscript to be more tightly centered on uncertainty quantification, in alignment with what the title suggests.
It is also noteworthy that the study demonstrates that, although TROPOMI observations provide a significant advancement, important observational gaps persist. The period from October to March remains essentially unobserved, suggesting that future missions could meaningfully support Arctic monitoring. Presenting the progress that can be achieved with the current observational record—while also highlighting the remaining potential—is highly valuable.
Below, I provide my comments section by section, indicating numbering at both general and detailed levels, and referring to specific lines when applicable:
0. Abstract: The abstract is concise and clearly presents the study. It effectively communicates the context, objectives, and main findings.
1. Introduction: The introduction appropriately presents the key elements that frame the research, identifies the gaps and challenges, and convincingly justifies the importance of studying the Arctic. Given that a substantial portion of the analysis focuses on tundra regions, I suggest including a land-cover map of the study domain, or alternatively a map distinguishing tundra and anthropogenic emission sources.
2.1 The work of Ward et al. (2024) is cited frequently throughout the manuscript. In several instances, the manuscript is not sufficiently self-contained, and the citation alone does not provide enough context for the reader to fully understand the methodological or conceptual point being referenced. While it is reasonable to avoid restating the entire content of Ward et al. (2024), the manuscript should nonetheless include the essential information needed for it to be understandable on its own.
Line 99: The methods section begins by citing Ward et al. (2024), followed by an explanation of the method. It is unclear whether the description corresponds directly to Ward et al. or whether it represents an adaptation or extension. A brief summary would help clarify this, as well as an explicit statement that further methodological details can be found in Ward et al. (2024).2.4 Figure 3, which presents the average footprints, is not referenced in the discussion. It would be valuable to connect the inversion results to the sensitivity patterns shown in this figure, particularly to assess how the inversions based on each product relate to sensitivities over tundra and anthropogenic regions. It would also be helpful to include the sensitivity of the BRW station in the figure.
2.6 The domain used for the inversions—particularly in the sensitivity tests—was not entirely clear. Figure 1 defines domain, region, and measurement sites; however, since multiple configurations are tested, it is important to explicitly clarify whether the same domain is consistently applied (if that is the case). It would also be helpful to explicitly state that the analysis focuses on the NSA region, even if this seems implied. Furthermore, to more directly assess uncertainty, it would have been interesting to include an experiment involving, for example, 10 perturbed prior ensembles and to evaluate the resulting spread.
In Figure 4b, the color scheme overlaps with that of Figure 1, which generated confusion. For instance, the Ward domain appears in orange in Figure 4a but in black in Figure 1.
I recommend relocating Figure 4 to the Results section, as it corresponds to the first point addressed there.
3.2.1 In this subsection, the inversions correspond to the NSA, but Alaska is used as the domain—is this correct?
3.2.2 – Figure 6 The shift between TROPOMI-OPER and BRW-based inversions is not addressed in either the results or the discussion. Although the manuscript comments on monthly matches, the phase differences are particularly striking and, in my view, merit discussion. This may even be connected to the sensitivity analysis previously conducted for TROPOMI.
3.3 – Figure 7 The legend lists “Uniform prior” twice. I assume that the solid line corresponds to the posterior of the Uniform-prior experiment and the dashed line to the prior.
3.3.4 – Lines 387–388 No Figures 9A or 9B are present.
3.6 To address uncertainty more directly, I recommend evaluating the spread of RMSE values for each TROPOMI product—that is, assessing how wide the RMSE range is across the experiments. The discussion currently emphasizes the RMSE magnitude for individual experiments and highlights improvements, but the variability among experiments is not addressed.
3.7 Tables S1 and S2 should be included in the main manuscript.
Line 435: The reference does not correspond to Figure S15; I understand that Figure S16 is intended instead.6. Conclusions Line 684: The recommendations from Tsuruta should be briefly summarized; including new citations in the conclusions is not advisable. The same applies to line 695.
Citation: https://doi.org/10.5194/egusphere-2025-4230-RC2 - AC2: 'Reply on RC2', Rebecca Ward, 16 Dec 2025
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,242 | 116 | 30 | 1,388 | 54 | 23 | 27 |
- HTML: 1,242
- PDF: 116
- XML: 30
- Total: 1,388
- Supplement: 54
- BibTeX: 23
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
1) General comments
Methane is second most important greenhouse gas in Earth's atmosphere,
and the Arctic region is a potential large source of CH4 emission.
Emission detection and monitoring is therefore an interesting and
relevant research question. The authors compare emission estimates
based on surface-only and TROPOMI based inversions using two different
retrieval products. A sensitivity study is conducted to asses the
robustness of the results. Since the article adresses a relevant
sicientific topic from an inversion modelling perspective, it
should be published after consideration of the specific comments
given below.
2) Specific comments
*) page 4, line 99: the inversion method used in this study is the same
as in Ward (2024). A few lines about the general setup of this method
would make the rest of the section easier to follow for those readers
who are not familiar with this method and don't have the time to read
the referenced paper (e.g. what is the "tuning phase" (line 115) or
"MCMC sample" (line 119), or how is the uncertainty interval (line 118)
calculated?).
*) page 5, line 108: in some models negative emissions are interpreted
as uptake instead. Why is that not considered here?
*) page 6, lines 156-158: I interpret regridding as calculating the
average of all observations falling into the gridcell of the
mentioned dimensions. This already reduces the data density of the
observations, so why is additional reduction step necessary? Would
using larger regridded "observations" without the additional
reduction step have the same effect as the current approach?
*) page 9 line 203-205: the XCH4^{model}_{pert} is apparently valid
from surface to maxlev (right-hand side of supplement eq 4 =
left-hand side of supplement eq. 7). That should be made clear in the
text as well, since now it's just "we have equation 2 and subtract
two other terms".
So these lines could be clarified by first stating why you would
need XCH4^{model}_{pert}, and then how you calculate it.
*) page 9, equation 4: if the first summation is from "1 till maxlev",
shouldn't the second one be from "maxlev+1 till n"? In other words,
which of the two summations includes maxlev? Related to that is
equation 3 and the rest of the derivation in the Supplement.
*) page 9, line 220: please clarify the CAMS dataset that you're using
and why you use an outdated version (based on the "data availability"
section, you could have downloaded one in 2023).
The problem with a "CAMS dataset" is that there are multiple datasets
providing methane estimates, see
https://ads.atmosphere.copernicus.eu/datasets?q=methane&limit=30
Neither of the datasets listed here is from 2019. The only dataset
that I found using a "v19r1" or similar versioning scheme is called
"CAMS global inversion-optimised greenhouse gas fluxes and concentrations"
for which the "Evaluation and quality assurance (EQA) reports" on the
documentation tab point to a report by Segers, not to Inness.
*) page 10, line 229: why use EDGAR v6? As you mention in section 4.1
(line 492-494), emission totals over Alaska may vary significantly
according to the version of the database being used.
3) Technical corrections
*) page 1, line 80: Since the acronym HBMCMC is used here, I would
capitalise the word "hierarchical" as well.
*) page 5, line 125: the 78W should be 78N?
*) page 6, line 156: please clarify if the regridding resolution is
latitude x longitude or the other way around? The same holds for
the other spatial resolutions mentioned hereafter.
*) Supplement page 1, eq. 6: the second summation term should have
n instead of 1 as it's upper level
*) page 19, line 387-388: there is no figure 9A or 9B.