the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Sep 2022
A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel and monsoon sources
Abstract. We interpret space-borne observations from the TROPOspheric Monitoring Instrument (TROPOMI) in a multi-inversion framework to characterize the 2018–2019 global methane budget. Evaluation of the inverse solutions indicates that methane sources and sinks cannot be separately resolved by methane observations alone—even with the dense TROPOMI sampling coverage. Employing remote carbon monoxide (CO) and hydroxyl radical (OH) observations as additional constraints, we infer from TROPOMI a global methane source of 587 Tg/y and sink of 571 Tg/y for our analysis period. We apply a new downscaling method to map these emissions to 0.1°×0.1° resolution, using the results to uncover key gaps in the prior methane budget. The TROPOMI data point to an underestimate of tropical wetland emissions (+13 %; 20 Tg/y), with adjustments following regional hydrology. Some simple wetland parameterizations represent these patterns as accurately as more sophisticated process-based models. Fossil fuel emissions are strongly underestimated over the Middle East (+5 Tg/y), where they have been increasing rapidly over the past decade, and over Venezuela. The TROPOMI observations reveal many fossil fuel emission hotspots missing from the prior inventory, including over Mexico, Oman, Yemen, Turkmenistan, Iran, Iraq, Libya, and Algeria. Agricultural methane sources are underestimated in India, Brazil, the California Central Valley, and Asia. More than 45 % of the global upward anthropogenic source adjustment occurs over India and southeast Asia during the summer monsoon (+8.5 Tg in Jul–Oct), likely due to rainfall-enhanced emissions from rice, manure, and landfills/sewers, which increase during this season along with the natural wetland source.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(3123 KB)
-
Supplement
(8014 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3123 KB) - Metadata XML
-
Supplement
(8014 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-948', John Worden, 25 Oct 2022
General comments (By John Worden).
The paper is mostly well written, except for the section comparing posterior OH to ATOM results. The main concern I have is that the method and results in this paper are not fundamentally different from those in the Qu et al. ACP 2021 paper which also uses TROPOMI data for quantifying emissions for essentially the same time frame. The main difference between this paper and the Qu et al. paper is the version of the data, which is ostensibly more accurate than the data used in Qu et al. but then there is no discussion on how this improved data set changes, or potentially improves the results over and above the Qu et al. results.
For acceptance, the paper needs to better describe the difference from those in Qu et al, and how these results are an improvement ; I think there are sufficient results in here for this purpose (e.g. monthly estimates allow for attributing some components of the methane budget). In addition, you could compare with the Qu et al. 2022 paper (methane surge) which uses GOSAT data for 2019; in principal the improved TROPOMI data sets should result in better comparisons with the GOSAT based results for this time period.
Another issue is a lack of discussion on uncertainties; I see them reported in final estimates but its not obvious how they are computed, are these buried in the text somewhere ?( Im pretty sure I read through the entire text, 2.5 times + browsing). A more extensive discussion on uncertainties should be in Section 2.
Note that Im not convinced that the downscaling approach described here is sufficient by itself to merit publication as it is not (obviously) an improvement over the optimal estimation based approach described in the un-cited Cusworth et al. 2021 paper (see subsequent comments).
Specific Comments:
Abstract, “… indicate rapid increases in Middle East”; the way this sentence is currently written implies you base this statement on satellite observations.
Abstract: state that you are estimating monthly values
Abstract: You stated you used observations of CO and OH, but I don’t see any description of these data in Section 2. Also see comments on comparisons to CO and OH to ATOM below
Section 2.1 Page 3: As stated in the general comments, the analogous paper here is from Qu et al. ACP 2021 which uses V1.03 TROPOMI data whereas you use the Lorente et al. based corrections; make that difference clear here. Note that as far as I can tell there is fundamentally no difference between yours and the Qu et al. results, notwithstanding the improved XCH4 data sets you are using.
Can you add discussion on this difference in Section 2.1 and then add more comparisons to the Qu et al. 2021 results in Section 4?
Section 2.6, page 7, Provide rationale for why you are downscaling to 0.1 degree resolution, especially since it depends on priors which can vary considerably (uncorrelated at 0.1 degree resolution) depending on choice of prior as you note in the text. As far as I can tell, the downscaled results are not used thereafter in the paper, is that correct? (Note that in the Worden et al. 2022 paper, we downscaled so that we can then upscale more accurately to each country; the other reason for the OE based downscaling (Cusworth et al. 2021) we developed is to step us towards using top-down emissions estimates for updating gridded inventories at this scale).
How does this downscaling approach compare to the optimal estimation based approach to downscaling discussed in Cusworth et al Earth Environ 2, 242 (2021). Can you perform a test(s) similar to what is shown in Cusworth et al. to ensure you are preserving information from original grid and downscaled grid? Your co-author A. Bloom designed these tests for Cusworth et al. so you could ask him for details. Note that I would be ecstatic for an additional vetting of this OE/Cusworth approach by the Dylan / Daven crew… we are pretty sure we got the math right as we used two different approaches to arrive at the same result (the Cusworth / Bloom and the Bowman approach, with Worden moderating), but given that its a 30+ equation derivation some additional vetting is desired.
Also cite Liu, M. et al. A New Divergence Method to Quantify Methane Emissions Using Observations of Sentinelâ5P TROPOMI. Geophys Res Lett 48, (2021), as a potential way to use satellite data to identify and quantify emissions at these same fine spatial scales.
Section 3.2. As a reader I did not understand either the rationale for the comparison to ATOM, or how I should interpret the comparison…. This section basically needs a re-write. Note that our group at JPL also attempted to use the ATOM OH estimates but decided against it (although this was a few years ago) because we did not have a good sense of the accuracy, especially since OH is tricky to measure; some discussion is needed on the ATOM OH accuracy to better interpret the comparison between your inversion results and these in situ results. Also, what did you intend to conclude from the comparison to CO?
Section 5.0, Compare agains the Ma et al. 2021 and Zhang et al. 2021 wetland results which suggest ~149 Tg CH4/yr total…this again might be a TROPOMI versus GOSAT issue as TROPOMI data results in lower livestock emissions than those from GOSAT in Brazil, which in turn would likely balance to the wetlands, relative to the GOSAT based results. A discussion here on these differences is needed.
Section 6, again compare these totals to the GOSAT based estimates (there are several now available). Discussion on potential TROPOMI / GOSAT differences are needed as well.
Section 6.2, Note that reports to UNFCC from Russia have varied considerably over the years, this should be discussed here (e.g. Scarpelli et al. 2021 versus Scarpelli et al. 2022).
7.0 Conclusions (and to some extent abstract). The paper implies that missing sources can be identified through the downscaling approach, but this is not possible if you are using prior emissions for the downscaling. Also, how can the Venezuelan source simultaneously be lower than the prior and inline with trend estimates? These are different quantities. I think you mean something else here.
7.0: Line 540 Conclusions about waste and agriculture priors being too small… yes we are finding this to be the case with all the other published TROPOMI and GOSAT based inversions, please reference these other papers.
7.0 Conclusions / Line 555: This conclusion is potentially very interesting but needs additional vetting. For one, how much of yearly Indian and Southeast Asian underestimate is due to the underestimate in the Monsoon seasons? In addition, how much of this is affected by smoothing error, which is not directly calculated using your method, but you could calculate by using different priors; basically we are finding significant impact of smoothing error, or alternatively cross-correlation of a change in one emission onto another, for emissions and their trends in this region.
References: You can peruse the Worden et al. ACP paper for missing references on GOSAT inversions that you can then compare to in the text; this same comment was made by reviewers of our Worden et al. paper.
Citation: https://doi.org/10.5194/egusphere-2022-948-RC1 -
AC1: 'Reply on RC1', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Xueying Yu, 17 Jan 2023
-
CC1: 'Comment on egusphere-2022-948', Alba Mols, 14 Nov 2022
14/11/2022 Review of “A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel and monsoon sources” by Yu et al.
This review was prepared as part of graduate program course work at Wageningen University, and has been produced under supervision of dr. Ingrid Luijkx. The review has been posted because of its good quality, and likely usefulness to the authors and editor. This review was not solicited by the journal.
The paper by Yu et, al. entitled “A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel and monsoon sources” presents a quantification of the 2018-2019 global methane budget, based on space-borne TROPOMI observations. Methane emissions are derived from the TROPOMI observations by coupling multiple 4D-Var adjoint inversions with a newly developed spatial downscaling approach. This enables the identification of previously missing or underestimated methane emissions from fossil fuel and wetland sources.
This research presents a new downscaling method that is applied to convert the GEOS-chem model output to a 0.1º × 0.1º resolution, using combined spatial information from the TROPOMI observations and from the prior estimates. This method enables very specific allocation of emission hotspots. In the study, OH is used as an additional constraint in the inversions, as recommended by Saunois et al. (2020). This advances previous studies, which often co-optimized methane sources and sinks by using methane data alone. The results section of the manuscript is well-written and discusses all the source areas in-depth, also suggesting possible underlying reasons for the found underestimations in prior inventories. The research reveals some interesting results regarding emission hotspots that were missing from prior inventories. That being said, I do have some remarks that could be addressed before publication.
1) If I understand it correctly, the aim of the research is to quantify the 2018-2019 global methane budget and determine missing and underrepresented emission sources. However, the authors mainly present how much the prior estimates are underestimated compared to their findings, making the results section an evaluation of their one specific chosen set of prior inventories. This approach results in a high dependency of the research on the choice of the specific prior estimates. If other prior inventories were chosen, the underestimations and hotspots that the research now revealed would likely be very different, because for instance, as was pointed out in the introduction, two of the most commonly used anthropogenic emission inventories (EDGAR v5 and GEPA) are uncorrelated at a 0.1º x 0.1º resolution. To overcome this issue, I would suggest to shift the focus of the results section from the discrepancies in the specific prior to the obtained absolute values of the methane budget. A good addition would then be a comparison of these results to independent measurements, such as the ObsPack or TCCON observations, or a comparison to other studies that also use inverse models to characterize the methane budget, such as Saunois et al. (2020).
2) The authors take an ensemble mean of the 4 different inversion formalisms to calculate the emission corrections, while their previous research showed that some of them perform better for different purposes (Yu et al., 2021a). The different allocation of emission hotspots that are found through the different inversions are already nicely discussed in the results section, but the emission corrections are subsequently still calculated as the multi-model mean. I would like to see a more in-depth discussion of why the authors chose this approach, and why for instance the classical SF inversion is not left out here, since it is highly biased towards areas where the prior estimates of the emissions are high, and therefore likely makes the calculated underestimations of the prior estimates too small (Yu et al., 2021a). The BI inversion approach provides the best spatial distribution of all inverse approaches, while the EE inversion performs best in recovering large missing sources (Yu et al., 2021a). In the calculations of the hotspot emissions that are missing from the prior inventories, it is therefore probably better to use the EE inversion instead of the ensemble mean. Table S1 could also be used in this discussion, since it summarizes the performance of the different inversion formalisms, while the statistics presented here are currently not used in the text.
3) The section of the development of the novel downscaling method could be more extensive. Since a new method is presented here, it’s very important that it is well-described. First of all, I would like to see the argumentation on why there was a need for a new downscaling method, and why previous downscaling methods were not suitable. M. Yu et al. (2021b) could be consulted, who present a nice review section of related work on spatial interpolation and downscaling of airborne pollutants. Also, since the downscaling method is presented as novel, a proper evaluation of its accuracy is very important. I therefore wonder why the authors chose to perform the OSSE only for one area, for the duration of one month and at a resolution of 0.25º x 0.3125º, and subsequently chose to use a 0.1º x 0.1º resolution in their further research based on this OSSE. The representativeness of this one OSSE for the whole research should be better discussed and possibly expanded, since the validity of the research is dependent on this outcome.
4) After the results section, I would suggest to include a section where the uncertainties in both the TROPOMI data and the prior estimates is discussed, since the research is very dependent on both, and therefore also dependent on errors in the data. Also, the methods could be further discussed in this section, such as implications of the downscaling of the optimized emissions, and the use of the different inversion formalisms.
5) In my view, the knowledge gap could be further specified in the introduction. The novel aspect of the methods is already highlighted well by stressing the importance of including OH constraints, which many previous studies did not include. However, a section on prior knowledge about hotspots and emission sources that are often underrepresented in prior estimates is missing, including how the research is still of added value to this. Hu et al. (2018), who used TROPOMI to map methane column concentrations for instance also observed the underestimated hotspot of the Sudd wetlands and Venezuela. Lu et al. (2021) performed an inversion study using GOSAT data and also revealed missing spots in observational data, but on a far coarser resolution than this study. I suppose that the authors mainly add to this because of the far higher resolution of the TROPOMI data they use, combined with the downscaling method, making it easier to pinpoint emissions to more specific locations.
6) The authors nicely present the main underrepresented sources and missing hotspots in the conclusion, but a section with the further implications of these findings is missing. In the last section of the conclusion, some recommendations for future research are given (lines 561-564), but the statements include no references confirming that the addition of datasets of CO, methyl chloroform and formaldehyde would indeed improve future inversions. Also, the novel downscaling method is not mentioned in the conclusion, while this method is probably also relevant for further research.
Minor comments
Title: The current title is appealing because it directly mentions the new findings, but in my view, it does not cover the whole scope and innovative aspect of the research. I would consider changing the title to something like: "A high-resolution global map of methane emissions inferred from an inversion of TROPOMI satellite data reveals missing emission hotspots and previously underestimated sources."
Line 46: For a better overview of the previous research, I would elaborate here on what the conflicting reasons are for methane increase apart from the emission increase over tropical regions, such as an increase in emissions in the energy sector, an increase in wetland emissions, and a decrease in mean OH (McNorton et al., 2018).
Line 89: Please include the Sentinel-5 precursor/TROPOMI Level 2 Product User Manual Methane as a reference for requiring quality filter > 0.5: https://sentinel.esa.int/documents/247904/2474726/Sentinel-5P-Level-2-Product-User-Manual-Methane.pdf/1808f165-0486-4840-ac1d-06194238fa96
Line 96: Apart from mentioning the slope, please report the R2 as well here as a measure for agreement (R2 = 0.67).
Line 117 - 128: Please elaborate on why these specific prior estimates are chosen, and perhaps also elaborate on how these datasets are constructed (by models/measurements)?
Line 118: Why did the authors chose to use the UNFCCC inventory from 2016? The new version from 2019 might be more representative for the study period.
Line 153: Please give a reference or explain why 50% uncertainty in the remaining sources is chosen.
Line 162: It would be good to explain here how the OH sensitivity study is exactly performed, and specifically state where in the formula of the cost function the different uncertainties are used.
Line 187: I wonder why the authors chose the values of 10% and 90% for the weight of the prior and the background respectively. Yu et al. (2021a) used 50% and 50% in their example of this background increment inversion formalism. Is this determined with sensitivity simulations similar as in the OG inversions? Please explain.
Line 271: “Our 2019-2018 … growth rate acceleration”: please elaborate on the implications of this statement on the findings that are presented in this paragraph.
Line 314: Could the authors further explain here why the locations in the boxes of figure 2b were chosen for the analysis? This is probably because TROPOMI observations differ from the prior estimates in these areas. But when looking at the map, I see that this is for instance also the case for northern Italy and the Southeast US. Why are these areas not discussed?
Line 317: If I understand it correctly, the average yearly source and sink values for the years 2018-2019 that are presented here are not based on two full yearly cycles. The timeframe of the analysis only spans from 05/2018 - 10/2019. However, figure 4 indicates that the sources and sinks show seasonal variation. To retrieve yearly average values for the sources and sinks, these values can’t be just averaged over a 1.5 yearly cycle. I would recommend to take these average values over one full yearly cycle, for instance from 10/2018-10/2019.
Line 412: I would move the explanation of figure 5c to line 396, since that is where the figure is first mentioned.
Line 442: Since these missing hotspots are one of the main outcomes of the research, the authors could consider to give their more exact locations, instead of only mentioning the countries.
Line 447: I wonder how the hotspots can be missing in the UNFCCC inventory and show up in the EIA, since it seems like the UNFCCC is based on the national activity data from the EIA (Scarpelli et al., 2020). Is this because the authors used the UNFCCC data from 2016, and these activities were maybe still unknown at that time? Please explain this here, or as I mentioned before, consider using the updated UNFCCC inventory from 2019.
Figure 4: Please consider to make figure 4a-d larger, since the dots are very hard to see. Figure 4e is currently not referred to in the text. Also, I wonder why only the FixOH emission is shown here, and not the loss. I would either remove the fixOH emission from this plot, or include the loss as well.
Figure 5: Figure 5a and 5c show information from previous research, while figure 5b shows main findings of the research. I would therefore suggest to make figure 5b a separate figure.
Figure S9: In my opinion, this figure could also be included in the main text, since it shows well how the outcomes of the four inversion formalism differ, and how the inversion ensemble is constructed.
Specific comments
Line 17: Please remove “CO” here, since CO is not used as a constraint.
Line 43: “the importance of” can be left out here.
Line 229: Write abbreviation of OSSE out in full.
Line 335: The total emissions of China mentioned here (60 Tg/y) is different from the number in table S2 (61 Tg/y). Please make this consistent.
Line 342: “Europe Union” > “European Union”.
Table 2: “Russian” > “Russia”.
References
Hu, H., Landgraf, J., Detmers, R., Borsdorff, T., Aan de Brugh, J., Aben, I., ... & Hasekamp, O. (2018). Toward global mapping of methane with TROPOMI: First results and intersatellite comparison to GOSAT. Geophysical Research Letters, 45(8), 3682-3689.
Lu, X., Jacob, D. J., Zhang, Y., Maasakkers, J. D., Sulprizio, M. P., Shen, L., ... & Ma, S. (2021). Global methane budget and trend, 2010–2017: complementarity of inverse analyses using in situ (GLOBALVIEWplus CH 4 ObsPack) and satellite (GOSAT) observations. Atmospheric Chemistry and Physics, 21(6), 4637-4657.
McNorton, J., Wilson, C., Gloor, M., Parker, R. J., Boesch, H., Feng, W., ... & Chipperfield, M. P. (2018). Attribution of recent increases in atmospheric methane through 3-D inverse modelling. Atmospheric Chemistry and Physics, 18(24), 18149-18168.
Saunois, M., Stavert, A. R., Poulter, B., Bousquet, P., Canadell, J. G., Jackson, R. B., ... & Zhuang, Q. (2020). The global methane budget 2000–2017. Earth system science data, 12(3), 1561-1623.
Scarpelli, T. R., Jacob, D. J., Maasakkers, J. D., Sulprizio, M. P., Sheng, J. X., Rose, K., ... & Janssens-Maenhout, G. (2020). A global gridded (0.1× 0.1) inventory of methane emissions from oil, gas, and coal exploitation based on national reports to the United Nations Framework Convention on Climate Change. Earth System Science Data, 12(1), 563-575.
Yu, X., Millet, D. B., & Henze, D. K. (2021a). How well can inverse analyses of high-resolution satellite data resolve heterogeneous methane fluxes? Observing system simulation experiments with the GEOS-Chem adjoint model (v35). Geoscientific Model Development, 14(12), 7775-7793.
Yu, M., & Liu, Q. (2021b). Deep learning-based downscaling of tropospheric nitrogen dioxide using ground-level and satellite observations. Science of The Total Environment, 773, 145145.
Citation: https://doi.org/10.5194/egusphere-2022-948-CC1 -
AC2: 'Reply on CC1', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC2-supplement.pdf
-
AC2: 'Reply on CC1', Xueying Yu, 17 Jan 2023
-
RC2: 'Comment on egusphere-2022-948', Anonymous Referee #2, 15 Nov 2022
The authors utilized the latest version of TROPOMI XCH4 retrievals averaged in 2018-2019 to optimize various sources of CH4 on a global scale. In addition, since OH and CH4 are intertwined, they added OH to the state vectors for adjustment (optOH). Finally, the authors proposed a statistical downscaling method leveraging both prior knowledge from bottom-up emission inventories and the oversampled TROPOMI data to scale the optimized 2x2.5 degree emissions down to 0.1x0.1 degree. This method enabled them to identify the missing sources better. Their primary take-home messages are i) The use of XCH4 observations is not adequate to provide reliable constraints on OH; this is why authors gave up on the optOH result; ii) the middle east underreports the energy-sector CH4 emissions; iii) In South and Southasia, there is a strong degree of correlation between CH4 agriculture and waste emissions and some hydrological variables such as more precipitation (or to be more precise, the runoff) during Monsoon seasons; iv) the reported emissions related to the oil and gas industry over the US in the latest bottom-emission inventory (EDGAR v6) is not too different than the top-down estimates made from this study. In general, the paper has important implications for the CH4 budget and regulations. However, the results are too optimistic because careful error quantification is lacking. In addition, some key aspects of inversions need to be clarified. A major revision is required to bring this draft to a publishable level.
Major comments:
Inversion: While I understand the importance of using an adjoint model for implicitly resolving the source-receptor relationship without having to rerun the forward model multiple times, the inversion framework comes with a significant weakness which is its inability to gauge the confidence level in the final estimates (i.e., the posterior error). The paper should inform the readers about this major weakness (introduction, conclusion, and Table 2) and highlights studies such as Qu et al. 2021, which reported the AKs of the top-down estimates because they used an analytical inversion. For example, can we trust the optimized CH4 emissions using the TROPOMI XCH4 over water or high SZA? are the reported top-down estimates statistically significant? In addition, several aspects of the inversion need to be further clarified: i) It needs to be explained how the 4D-var framework is applied when the inversion window is as wide as a 2-years average. ii) the TROPOMI full-physics algorithm relies on the prior profiles meaning the retrieved XCH4 is a piece of information on top of an ignorant model; I do not see any mention of if TROPOMI XCH4 was recalculated with GEOS-Chem prior profiles to ensure that only the true information from the satellite radiance is used for the inversion (Page 39 in http://www.tropomi.eu/sites/default/files/files/publicSentinel-5P-TROPOMI-ATBD-Methane-retrieval.pdf). This task should be done iteratively because the GEOS-Chem profiles change after each inversion iteration. iii) it is unclear if the TROPOMI data have been scaled up to the resolution of GEOS-Chem in the inversion; if not, the difference in their spatial representativity will result in a perceived bias which can be problematic. iv) how do the errors associated with the vertical diffusion in GEOS-Chem impact your result? The model error parameter is lacking in the analysis. v) is the state vector the total CH4 emissions, or is it sector-based? vi) why did not the authors use the glint mode to account for off-shore emissions?
Downscaling: Two central problems exist: i) can the two-year average TROPOMI XCH4 truly capture the spatial variance in XCH4 at 0.1x0.1 degrees? By oversampling TROPOMI pixels, we may lose spatial variance (a smoothing effect) more than we reduce the random noises. The authors need to prove that a two-year averaged TROPOMI data can resolve the length scales of plumes at the resolution of 0.1x0.1 degree; if not, that resulting spatial representativity error induced by oversampling can potentially hinder reaching a 0.1x0.1 degree information. ii) The proposed downscaling method (Eq2) heavily relies on assumptions about prior errors/information. Even the observational term depends on the prior fraction of emissions (fk). As noted by the authors, the prior CH4 emissions do not agree with other bottom-up emission inventories (R2=0.01?), so how can one fully trust a downscaling output when it heavily relies on questionable prior information? Would it be more sensible to use the posterior error/distribution for this part (under the condition in which the inversion framework was analytical, permitting the calculation of the posterior error)? As a result of these two combined complications, I challenge the authors to provide an error estimation for this downscaling method and propagate them to the emissions maps and statistics (especially Table 2). Your study did not inform the posterior errors due to the use of adjoint; now, the lack of an uncertainty estimation for the downscaling part (which is the most crucial selling point of the paper) appears as an oversight.
Comparison to ATOM: I need clarification on this comparison. Based on the author's discussion on optOH, they suggested that a strong El Nino year (2018-2019) led to lower-than-average OH mixing ratio (8.27e5 molec/cm3). But then they compared their constrained model in different years (<2018) with ATHOS OH measurements and concluded that their optOH is vastly underestimated, reinforced by the overestimation of CO. If 2018-2019 was a unique timeframe, how could one generalize the comparison results from other years to the 2018-2019 period? Furthermore, I see a few issues here i) ATHOS OH can easily contain up to 30% error; have the authors considered the measurement errors in their comparison? Given the observational errors, I encourage applying a statistical test to know if the differences are real. ii) have the authors looked into the measured OHR to see if there are missing sources (such as VOC) in their model? What is the implication of underestimating OH in the optOH scenario? Should we perform a multispecies inversion with TROPOMI CO and HCHO to provide an additional constraint on OH? In theory, letting OH see the CH4 feedback (optOH) is suitable, but why should the ATOM analysis discourage this meaningful practice? What if your default CO simulations are too high, and the optOH highlights that tendency? I'm left with many questions because the authors needed to dig into the problem more deeply.
Specific comments:
P1. L16. I do not think you ever used CO as an observational constraint for the model. This sentence is misleading.
P2. L38. Does really the recent enhancement in CH4 need to be better understood? I suggest adding more recent studies discussing the role of reduced NOx due to the lockdown on OH and CH4. There must be a recent study from Jacob's group regarding the increases in wetlands and permafrost CH4. This part needs more references in general.
P2. L40. After reading the abstract saying that sinks and sources cannot be resolved with a high-resolution satellite, I found this sentence regarding transformative advancement somewhat contradictory.
P2. L43. Shouldn't we also have an overrepresented source? If a source is underrepresented, another source should compensate for it.
P2. I found the second paragraph of the introduction imbalanced. The paper utilized remote sensing data, so I highly suggest comparing the pros and cons of using different remote sensing observations.
P2. L60. Who came up with this R2 value? The agreement is unsettlingly low. Please provide a reference.
P2. The third paragraph needs to include the temporal representation error between different emission inventories and the fact that CH4 emitters can vary from time to time at a relatively short temporal scale.
P3. L71. In the abstract, you said you had constrained CO, but here you imply that they will be used for evaluation.
P5. L147. Why is the regularization factor applied to So instead of Se? We are less confident in Se compared to So. Another way (which should be the same as finding the maximum curvature in the L-curve) to find the optimum regularization factor is to scale Se several times and find the knee point in averaging kernels vs. the factor, although this might not be possible with the adjoint.
P6. L167. How sure are you that the muted response of the model to a higher error in OH is not due to the lack of the degree of freedom? An analytical inversion would be able to answer it.
P6. L177. What is the implication of this low gamma value? The prior error is too uncertain or the observations are less noisy compared to the So?
P7. L213. Why should we accept this ad-hoc definition as XCH4 background? Any concerete evidence?
P9. Section 3.1. How can the errors in the soil uptake by methanotroph influence these results?
P10. In the first paragraph, I encourage using absolute numbers from Figure 3 to describe the reduction in bias, such as (from -13.8 to 8.8 ppbv).
P10. L299-301. I'm afraid I have to disagree with saying that TROPOMI is dense, but we cannot fully resolve the source/sink of CH4. It would help if you had more than CH4 to get OH right (such as HCHO and CO constraints), which has nothing to do with the densityTROPOMI XCH4 observations.
P10. In the second paragraph, you should specifically mention what emissions are used for the retrospective simulations.
P11. Is rice cultivation part of the wetland?
P12. Why talk about livestock in the wetland sections?
P12. 380. Where is the South Sudd in the figure?
P13. Can you provide more physical explanations of why these wetland emission models disagree so much? Is it due to their parameterization or the need for more information about water nitrogen content, heat content, depth of wetland, sulfate content, etc.?
P14. The second paragraph: are we so clueless about the wetland anaerobic activity to use a simple correlation analysis? How about the soil nitrogen, water temperature, depth, oxygen content, etc.?
P14. L421. Please add a fraction of the total for each sector.
P15. L444. Are they missing from other top-down emissions too? The bar is usually low for bottom-up emission inventories, especially in developing countries.
P17. L 515. Does correlation explain causation?
P18. L 560. It's not about the density of TROPOMI data but a piece of factual information from XCH4. We need more compounds not denser data.
P16. is this number of available pixels really a lot? Please provide the percentage for a hypothetical situation when clouds were not present.
Editorial comments:
P1. L16. What do you mean by separately resolved?
P1. L20-21. It is vague; does the hydrological adjustment come after or before?
P1. L22. The sentence (Fossil fuel emission...) is awkward.
P1. L23. Many -> several
P2. L39. What do you mean by strong heterogeneity? Spatial or temporal?
P2. L42. inverse -> inversion
P2. Please use +- for a normal range. 18+-1 is shorter and neater. Please apply this to the entire manuscript.
P8. L242. What do you mean by "spatial source uncertainty"?
P7. Eq.2. The i->j is weird; what do you mean?
Figure 4. The panels are too small.
Figure 5 needs to be enlarged. This is the most critical figure, which is hard to see.
Table2. The numbers in the parenthesis are just the deviation in a defined box, not an actual error. Please inform the readers about it.
Citation: https://doi.org/10.5194/egusphere-2022-948-RC2 -
AC3: 'Reply on RC2', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC3-supplement.pdf
-
AC3: 'Reply on RC2', Xueying Yu, 17 Jan 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2022-948', John Worden, 25 Oct 2022
General comments (By John Worden).
The paper is mostly well written, except for the section comparing posterior OH to ATOM results. The main concern I have is that the method and results in this paper are not fundamentally different from those in the Qu et al. ACP 2021 paper which also uses TROPOMI data for quantifying emissions for essentially the same time frame. The main difference between this paper and the Qu et al. paper is the version of the data, which is ostensibly more accurate than the data used in Qu et al. but then there is no discussion on how this improved data set changes, or potentially improves the results over and above the Qu et al. results.
For acceptance, the paper needs to better describe the difference from those in Qu et al, and how these results are an improvement ; I think there are sufficient results in here for this purpose (e.g. monthly estimates allow for attributing some components of the methane budget). In addition, you could compare with the Qu et al. 2022 paper (methane surge) which uses GOSAT data for 2019; in principal the improved TROPOMI data sets should result in better comparisons with the GOSAT based results for this time period.
Another issue is a lack of discussion on uncertainties; I see them reported in final estimates but its not obvious how they are computed, are these buried in the text somewhere ?( Im pretty sure I read through the entire text, 2.5 times + browsing). A more extensive discussion on uncertainties should be in Section 2.
Note that Im not convinced that the downscaling approach described here is sufficient by itself to merit publication as it is not (obviously) an improvement over the optimal estimation based approach described in the un-cited Cusworth et al. 2021 paper (see subsequent comments).
Specific Comments:
Abstract, “… indicate rapid increases in Middle East”; the way this sentence is currently written implies you base this statement on satellite observations.
Abstract: state that you are estimating monthly values
Abstract: You stated you used observations of CO and OH, but I don’t see any description of these data in Section 2. Also see comments on comparisons to CO and OH to ATOM below
Section 2.1 Page 3: As stated in the general comments, the analogous paper here is from Qu et al. ACP 2021 which uses V1.03 TROPOMI data whereas you use the Lorente et al. based corrections; make that difference clear here. Note that as far as I can tell there is fundamentally no difference between yours and the Qu et al. results, notwithstanding the improved XCH4 data sets you are using.
Can you add discussion on this difference in Section 2.1 and then add more comparisons to the Qu et al. 2021 results in Section 4?
Section 2.6, page 7, Provide rationale for why you are downscaling to 0.1 degree resolution, especially since it depends on priors which can vary considerably (uncorrelated at 0.1 degree resolution) depending on choice of prior as you note in the text. As far as I can tell, the downscaled results are not used thereafter in the paper, is that correct? (Note that in the Worden et al. 2022 paper, we downscaled so that we can then upscale more accurately to each country; the other reason for the OE based downscaling (Cusworth et al. 2021) we developed is to step us towards using top-down emissions estimates for updating gridded inventories at this scale).
How does this downscaling approach compare to the optimal estimation based approach to downscaling discussed in Cusworth et al Earth Environ 2, 242 (2021). Can you perform a test(s) similar to what is shown in Cusworth et al. to ensure you are preserving information from original grid and downscaled grid? Your co-author A. Bloom designed these tests for Cusworth et al. so you could ask him for details. Note that I would be ecstatic for an additional vetting of this OE/Cusworth approach by the Dylan / Daven crew… we are pretty sure we got the math right as we used two different approaches to arrive at the same result (the Cusworth / Bloom and the Bowman approach, with Worden moderating), but given that its a 30+ equation derivation some additional vetting is desired.
Also cite Liu, M. et al. A New Divergence Method to Quantify Methane Emissions Using Observations of Sentinelâ5P TROPOMI. Geophys Res Lett 48, (2021), as a potential way to use satellite data to identify and quantify emissions at these same fine spatial scales.
Section 3.2. As a reader I did not understand either the rationale for the comparison to ATOM, or how I should interpret the comparison…. This section basically needs a re-write. Note that our group at JPL also attempted to use the ATOM OH estimates but decided against it (although this was a few years ago) because we did not have a good sense of the accuracy, especially since OH is tricky to measure; some discussion is needed on the ATOM OH accuracy to better interpret the comparison between your inversion results and these in situ results. Also, what did you intend to conclude from the comparison to CO?
Section 5.0, Compare agains the Ma et al. 2021 and Zhang et al. 2021 wetland results which suggest ~149 Tg CH4/yr total…this again might be a TROPOMI versus GOSAT issue as TROPOMI data results in lower livestock emissions than those from GOSAT in Brazil, which in turn would likely balance to the wetlands, relative to the GOSAT based results. A discussion here on these differences is needed.
Section 6, again compare these totals to the GOSAT based estimates (there are several now available). Discussion on potential TROPOMI / GOSAT differences are needed as well.
Section 6.2, Note that reports to UNFCC from Russia have varied considerably over the years, this should be discussed here (e.g. Scarpelli et al. 2021 versus Scarpelli et al. 2022).
7.0 Conclusions (and to some extent abstract). The paper implies that missing sources can be identified through the downscaling approach, but this is not possible if you are using prior emissions for the downscaling. Also, how can the Venezuelan source simultaneously be lower than the prior and inline with trend estimates? These are different quantities. I think you mean something else here.
7.0: Line 540 Conclusions about waste and agriculture priors being too small… yes we are finding this to be the case with all the other published TROPOMI and GOSAT based inversions, please reference these other papers.
7.0 Conclusions / Line 555: This conclusion is potentially very interesting but needs additional vetting. For one, how much of yearly Indian and Southeast Asian underestimate is due to the underestimate in the Monsoon seasons? In addition, how much of this is affected by smoothing error, which is not directly calculated using your method, but you could calculate by using different priors; basically we are finding significant impact of smoothing error, or alternatively cross-correlation of a change in one emission onto another, for emissions and their trends in this region.
References: You can peruse the Worden et al. ACP paper for missing references on GOSAT inversions that you can then compare to in the text; this same comment was made by reviewers of our Worden et al. paper.
Citation: https://doi.org/10.5194/egusphere-2022-948-RC1 -
AC1: 'Reply on RC1', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Xueying Yu, 17 Jan 2023
-
CC1: 'Comment on egusphere-2022-948', Alba Mols, 14 Nov 2022
14/11/2022 Review of “A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel and monsoon sources” by Yu et al.
This review was prepared as part of graduate program course work at Wageningen University, and has been produced under supervision of dr. Ingrid Luijkx. The review has been posted because of its good quality, and likely usefulness to the authors and editor. This review was not solicited by the journal.
The paper by Yu et, al. entitled “A high-resolution satellite-based map of global methane emissions reveals missing wetland, fossil fuel and monsoon sources” presents a quantification of the 2018-2019 global methane budget, based on space-borne TROPOMI observations. Methane emissions are derived from the TROPOMI observations by coupling multiple 4D-Var adjoint inversions with a newly developed spatial downscaling approach. This enables the identification of previously missing or underestimated methane emissions from fossil fuel and wetland sources.
This research presents a new downscaling method that is applied to convert the GEOS-chem model output to a 0.1º × 0.1º resolution, using combined spatial information from the TROPOMI observations and from the prior estimates. This method enables very specific allocation of emission hotspots. In the study, OH is used as an additional constraint in the inversions, as recommended by Saunois et al. (2020). This advances previous studies, which often co-optimized methane sources and sinks by using methane data alone. The results section of the manuscript is well-written and discusses all the source areas in-depth, also suggesting possible underlying reasons for the found underestimations in prior inventories. The research reveals some interesting results regarding emission hotspots that were missing from prior inventories. That being said, I do have some remarks that could be addressed before publication.
1) If I understand it correctly, the aim of the research is to quantify the 2018-2019 global methane budget and determine missing and underrepresented emission sources. However, the authors mainly present how much the prior estimates are underestimated compared to their findings, making the results section an evaluation of their one specific chosen set of prior inventories. This approach results in a high dependency of the research on the choice of the specific prior estimates. If other prior inventories were chosen, the underestimations and hotspots that the research now revealed would likely be very different, because for instance, as was pointed out in the introduction, two of the most commonly used anthropogenic emission inventories (EDGAR v5 and GEPA) are uncorrelated at a 0.1º x 0.1º resolution. To overcome this issue, I would suggest to shift the focus of the results section from the discrepancies in the specific prior to the obtained absolute values of the methane budget. A good addition would then be a comparison of these results to independent measurements, such as the ObsPack or TCCON observations, or a comparison to other studies that also use inverse models to characterize the methane budget, such as Saunois et al. (2020).
2) The authors take an ensemble mean of the 4 different inversion formalisms to calculate the emission corrections, while their previous research showed that some of them perform better for different purposes (Yu et al., 2021a). The different allocation of emission hotspots that are found through the different inversions are already nicely discussed in the results section, but the emission corrections are subsequently still calculated as the multi-model mean. I would like to see a more in-depth discussion of why the authors chose this approach, and why for instance the classical SF inversion is not left out here, since it is highly biased towards areas where the prior estimates of the emissions are high, and therefore likely makes the calculated underestimations of the prior estimates too small (Yu et al., 2021a). The BI inversion approach provides the best spatial distribution of all inverse approaches, while the EE inversion performs best in recovering large missing sources (Yu et al., 2021a). In the calculations of the hotspot emissions that are missing from the prior inventories, it is therefore probably better to use the EE inversion instead of the ensemble mean. Table S1 could also be used in this discussion, since it summarizes the performance of the different inversion formalisms, while the statistics presented here are currently not used in the text.
3) The section of the development of the novel downscaling method could be more extensive. Since a new method is presented here, it’s very important that it is well-described. First of all, I would like to see the argumentation on why there was a need for a new downscaling method, and why previous downscaling methods were not suitable. M. Yu et al. (2021b) could be consulted, who present a nice review section of related work on spatial interpolation and downscaling of airborne pollutants. Also, since the downscaling method is presented as novel, a proper evaluation of its accuracy is very important. I therefore wonder why the authors chose to perform the OSSE only for one area, for the duration of one month and at a resolution of 0.25º x 0.3125º, and subsequently chose to use a 0.1º x 0.1º resolution in their further research based on this OSSE. The representativeness of this one OSSE for the whole research should be better discussed and possibly expanded, since the validity of the research is dependent on this outcome.
4) After the results section, I would suggest to include a section where the uncertainties in both the TROPOMI data and the prior estimates is discussed, since the research is very dependent on both, and therefore also dependent on errors in the data. Also, the methods could be further discussed in this section, such as implications of the downscaling of the optimized emissions, and the use of the different inversion formalisms.
5) In my view, the knowledge gap could be further specified in the introduction. The novel aspect of the methods is already highlighted well by stressing the importance of including OH constraints, which many previous studies did not include. However, a section on prior knowledge about hotspots and emission sources that are often underrepresented in prior estimates is missing, including how the research is still of added value to this. Hu et al. (2018), who used TROPOMI to map methane column concentrations for instance also observed the underestimated hotspot of the Sudd wetlands and Venezuela. Lu et al. (2021) performed an inversion study using GOSAT data and also revealed missing spots in observational data, but on a far coarser resolution than this study. I suppose that the authors mainly add to this because of the far higher resolution of the TROPOMI data they use, combined with the downscaling method, making it easier to pinpoint emissions to more specific locations.
6) The authors nicely present the main underrepresented sources and missing hotspots in the conclusion, but a section with the further implications of these findings is missing. In the last section of the conclusion, some recommendations for future research are given (lines 561-564), but the statements include no references confirming that the addition of datasets of CO, methyl chloroform and formaldehyde would indeed improve future inversions. Also, the novel downscaling method is not mentioned in the conclusion, while this method is probably also relevant for further research.
Minor comments
Title: The current title is appealing because it directly mentions the new findings, but in my view, it does not cover the whole scope and innovative aspect of the research. I would consider changing the title to something like: "A high-resolution global map of methane emissions inferred from an inversion of TROPOMI satellite data reveals missing emission hotspots and previously underestimated sources."
Line 46: For a better overview of the previous research, I would elaborate here on what the conflicting reasons are for methane increase apart from the emission increase over tropical regions, such as an increase in emissions in the energy sector, an increase in wetland emissions, and a decrease in mean OH (McNorton et al., 2018).
Line 89: Please include the Sentinel-5 precursor/TROPOMI Level 2 Product User Manual Methane as a reference for requiring quality filter > 0.5: https://sentinel.esa.int/documents/247904/2474726/Sentinel-5P-Level-2-Product-User-Manual-Methane.pdf/1808f165-0486-4840-ac1d-06194238fa96
Line 96: Apart from mentioning the slope, please report the R2 as well here as a measure for agreement (R2 = 0.67).
Line 117 - 128: Please elaborate on why these specific prior estimates are chosen, and perhaps also elaborate on how these datasets are constructed (by models/measurements)?
Line 118: Why did the authors chose to use the UNFCCC inventory from 2016? The new version from 2019 might be more representative for the study period.
Line 153: Please give a reference or explain why 50% uncertainty in the remaining sources is chosen.
Line 162: It would be good to explain here how the OH sensitivity study is exactly performed, and specifically state where in the formula of the cost function the different uncertainties are used.
Line 187: I wonder why the authors chose the values of 10% and 90% for the weight of the prior and the background respectively. Yu et al. (2021a) used 50% and 50% in their example of this background increment inversion formalism. Is this determined with sensitivity simulations similar as in the OG inversions? Please explain.
Line 271: “Our 2019-2018 … growth rate acceleration”: please elaborate on the implications of this statement on the findings that are presented in this paragraph.
Line 314: Could the authors further explain here why the locations in the boxes of figure 2b were chosen for the analysis? This is probably because TROPOMI observations differ from the prior estimates in these areas. But when looking at the map, I see that this is for instance also the case for northern Italy and the Southeast US. Why are these areas not discussed?
Line 317: If I understand it correctly, the average yearly source and sink values for the years 2018-2019 that are presented here are not based on two full yearly cycles. The timeframe of the analysis only spans from 05/2018 - 10/2019. However, figure 4 indicates that the sources and sinks show seasonal variation. To retrieve yearly average values for the sources and sinks, these values can’t be just averaged over a 1.5 yearly cycle. I would recommend to take these average values over one full yearly cycle, for instance from 10/2018-10/2019.
Line 412: I would move the explanation of figure 5c to line 396, since that is where the figure is first mentioned.
Line 442: Since these missing hotspots are one of the main outcomes of the research, the authors could consider to give their more exact locations, instead of only mentioning the countries.
Line 447: I wonder how the hotspots can be missing in the UNFCCC inventory and show up in the EIA, since it seems like the UNFCCC is based on the national activity data from the EIA (Scarpelli et al., 2020). Is this because the authors used the UNFCCC data from 2016, and these activities were maybe still unknown at that time? Please explain this here, or as I mentioned before, consider using the updated UNFCCC inventory from 2019.
Figure 4: Please consider to make figure 4a-d larger, since the dots are very hard to see. Figure 4e is currently not referred to in the text. Also, I wonder why only the FixOH emission is shown here, and not the loss. I would either remove the fixOH emission from this plot, or include the loss as well.
Figure 5: Figure 5a and 5c show information from previous research, while figure 5b shows main findings of the research. I would therefore suggest to make figure 5b a separate figure.
Figure S9: In my opinion, this figure could also be included in the main text, since it shows well how the outcomes of the four inversion formalism differ, and how the inversion ensemble is constructed.
Specific comments
Line 17: Please remove “CO” here, since CO is not used as a constraint.
Line 43: “the importance of” can be left out here.
Line 229: Write abbreviation of OSSE out in full.
Line 335: The total emissions of China mentioned here (60 Tg/y) is different from the number in table S2 (61 Tg/y). Please make this consistent.
Line 342: “Europe Union” > “European Union”.
Table 2: “Russian” > “Russia”.
References
Hu, H., Landgraf, J., Detmers, R., Borsdorff, T., Aan de Brugh, J., Aben, I., ... & Hasekamp, O. (2018). Toward global mapping of methane with TROPOMI: First results and intersatellite comparison to GOSAT. Geophysical Research Letters, 45(8), 3682-3689.
Lu, X., Jacob, D. J., Zhang, Y., Maasakkers, J. D., Sulprizio, M. P., Shen, L., ... & Ma, S. (2021). Global methane budget and trend, 2010–2017: complementarity of inverse analyses using in situ (GLOBALVIEWplus CH 4 ObsPack) and satellite (GOSAT) observations. Atmospheric Chemistry and Physics, 21(6), 4637-4657.
McNorton, J., Wilson, C., Gloor, M., Parker, R. J., Boesch, H., Feng, W., ... & Chipperfield, M. P. (2018). Attribution of recent increases in atmospheric methane through 3-D inverse modelling. Atmospheric Chemistry and Physics, 18(24), 18149-18168.
Saunois, M., Stavert, A. R., Poulter, B., Bousquet, P., Canadell, J. G., Jackson, R. B., ... & Zhuang, Q. (2020). The global methane budget 2000–2017. Earth system science data, 12(3), 1561-1623.
Scarpelli, T. R., Jacob, D. J., Maasakkers, J. D., Sulprizio, M. P., Sheng, J. X., Rose, K., ... & Janssens-Maenhout, G. (2020). A global gridded (0.1× 0.1) inventory of methane emissions from oil, gas, and coal exploitation based on national reports to the United Nations Framework Convention on Climate Change. Earth System Science Data, 12(1), 563-575.
Yu, X., Millet, D. B., & Henze, D. K. (2021a). How well can inverse analyses of high-resolution satellite data resolve heterogeneous methane fluxes? Observing system simulation experiments with the GEOS-Chem adjoint model (v35). Geoscientific Model Development, 14(12), 7775-7793.
Yu, M., & Liu, Q. (2021b). Deep learning-based downscaling of tropospheric nitrogen dioxide using ground-level and satellite observations. Science of The Total Environment, 773, 145145.
Citation: https://doi.org/10.5194/egusphere-2022-948-CC1 -
AC2: 'Reply on CC1', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC2-supplement.pdf
-
AC2: 'Reply on CC1', Xueying Yu, 17 Jan 2023
-
RC2: 'Comment on egusphere-2022-948', Anonymous Referee #2, 15 Nov 2022
The authors utilized the latest version of TROPOMI XCH4 retrievals averaged in 2018-2019 to optimize various sources of CH4 on a global scale. In addition, since OH and CH4 are intertwined, they added OH to the state vectors for adjustment (optOH). Finally, the authors proposed a statistical downscaling method leveraging both prior knowledge from bottom-up emission inventories and the oversampled TROPOMI data to scale the optimized 2x2.5 degree emissions down to 0.1x0.1 degree. This method enabled them to identify the missing sources better. Their primary take-home messages are i) The use of XCH4 observations is not adequate to provide reliable constraints on OH; this is why authors gave up on the optOH result; ii) the middle east underreports the energy-sector CH4 emissions; iii) In South and Southasia, there is a strong degree of correlation between CH4 agriculture and waste emissions and some hydrological variables such as more precipitation (or to be more precise, the runoff) during Monsoon seasons; iv) the reported emissions related to the oil and gas industry over the US in the latest bottom-emission inventory (EDGAR v6) is not too different than the top-down estimates made from this study. In general, the paper has important implications for the CH4 budget and regulations. However, the results are too optimistic because careful error quantification is lacking. In addition, some key aspects of inversions need to be clarified. A major revision is required to bring this draft to a publishable level.
Major comments:
Inversion: While I understand the importance of using an adjoint model for implicitly resolving the source-receptor relationship without having to rerun the forward model multiple times, the inversion framework comes with a significant weakness which is its inability to gauge the confidence level in the final estimates (i.e., the posterior error). The paper should inform the readers about this major weakness (introduction, conclusion, and Table 2) and highlights studies such as Qu et al. 2021, which reported the AKs of the top-down estimates because they used an analytical inversion. For example, can we trust the optimized CH4 emissions using the TROPOMI XCH4 over water or high SZA? are the reported top-down estimates statistically significant? In addition, several aspects of the inversion need to be further clarified: i) It needs to be explained how the 4D-var framework is applied when the inversion window is as wide as a 2-years average. ii) the TROPOMI full-physics algorithm relies on the prior profiles meaning the retrieved XCH4 is a piece of information on top of an ignorant model; I do not see any mention of if TROPOMI XCH4 was recalculated with GEOS-Chem prior profiles to ensure that only the true information from the satellite radiance is used for the inversion (Page 39 in http://www.tropomi.eu/sites/default/files/files/publicSentinel-5P-TROPOMI-ATBD-Methane-retrieval.pdf). This task should be done iteratively because the GEOS-Chem profiles change after each inversion iteration. iii) it is unclear if the TROPOMI data have been scaled up to the resolution of GEOS-Chem in the inversion; if not, the difference in their spatial representativity will result in a perceived bias which can be problematic. iv) how do the errors associated with the vertical diffusion in GEOS-Chem impact your result? The model error parameter is lacking in the analysis. v) is the state vector the total CH4 emissions, or is it sector-based? vi) why did not the authors use the glint mode to account for off-shore emissions?
Downscaling: Two central problems exist: i) can the two-year average TROPOMI XCH4 truly capture the spatial variance in XCH4 at 0.1x0.1 degrees? By oversampling TROPOMI pixels, we may lose spatial variance (a smoothing effect) more than we reduce the random noises. The authors need to prove that a two-year averaged TROPOMI data can resolve the length scales of plumes at the resolution of 0.1x0.1 degree; if not, that resulting spatial representativity error induced by oversampling can potentially hinder reaching a 0.1x0.1 degree information. ii) The proposed downscaling method (Eq2) heavily relies on assumptions about prior errors/information. Even the observational term depends on the prior fraction of emissions (fk). As noted by the authors, the prior CH4 emissions do not agree with other bottom-up emission inventories (R2=0.01?), so how can one fully trust a downscaling output when it heavily relies on questionable prior information? Would it be more sensible to use the posterior error/distribution for this part (under the condition in which the inversion framework was analytical, permitting the calculation of the posterior error)? As a result of these two combined complications, I challenge the authors to provide an error estimation for this downscaling method and propagate them to the emissions maps and statistics (especially Table 2). Your study did not inform the posterior errors due to the use of adjoint; now, the lack of an uncertainty estimation for the downscaling part (which is the most crucial selling point of the paper) appears as an oversight.
Comparison to ATOM: I need clarification on this comparison. Based on the author's discussion on optOH, they suggested that a strong El Nino year (2018-2019) led to lower-than-average OH mixing ratio (8.27e5 molec/cm3). But then they compared their constrained model in different years (<2018) with ATHOS OH measurements and concluded that their optOH is vastly underestimated, reinforced by the overestimation of CO. If 2018-2019 was a unique timeframe, how could one generalize the comparison results from other years to the 2018-2019 period? Furthermore, I see a few issues here i) ATHOS OH can easily contain up to 30% error; have the authors considered the measurement errors in their comparison? Given the observational errors, I encourage applying a statistical test to know if the differences are real. ii) have the authors looked into the measured OHR to see if there are missing sources (such as VOC) in their model? What is the implication of underestimating OH in the optOH scenario? Should we perform a multispecies inversion with TROPOMI CO and HCHO to provide an additional constraint on OH? In theory, letting OH see the CH4 feedback (optOH) is suitable, but why should the ATOM analysis discourage this meaningful practice? What if your default CO simulations are too high, and the optOH highlights that tendency? I'm left with many questions because the authors needed to dig into the problem more deeply.
Specific comments:
P1. L16. I do not think you ever used CO as an observational constraint for the model. This sentence is misleading.
P2. L38. Does really the recent enhancement in CH4 need to be better understood? I suggest adding more recent studies discussing the role of reduced NOx due to the lockdown on OH and CH4. There must be a recent study from Jacob's group regarding the increases in wetlands and permafrost CH4. This part needs more references in general.
P2. L40. After reading the abstract saying that sinks and sources cannot be resolved with a high-resolution satellite, I found this sentence regarding transformative advancement somewhat contradictory.
P2. L43. Shouldn't we also have an overrepresented source? If a source is underrepresented, another source should compensate for it.
P2. I found the second paragraph of the introduction imbalanced. The paper utilized remote sensing data, so I highly suggest comparing the pros and cons of using different remote sensing observations.
P2. L60. Who came up with this R2 value? The agreement is unsettlingly low. Please provide a reference.
P2. The third paragraph needs to include the temporal representation error between different emission inventories and the fact that CH4 emitters can vary from time to time at a relatively short temporal scale.
P3. L71. In the abstract, you said you had constrained CO, but here you imply that they will be used for evaluation.
P5. L147. Why is the regularization factor applied to So instead of Se? We are less confident in Se compared to So. Another way (which should be the same as finding the maximum curvature in the L-curve) to find the optimum regularization factor is to scale Se several times and find the knee point in averaging kernels vs. the factor, although this might not be possible with the adjoint.
P6. L167. How sure are you that the muted response of the model to a higher error in OH is not due to the lack of the degree of freedom? An analytical inversion would be able to answer it.
P6. L177. What is the implication of this low gamma value? The prior error is too uncertain or the observations are less noisy compared to the So?
P7. L213. Why should we accept this ad-hoc definition as XCH4 background? Any concerete evidence?
P9. Section 3.1. How can the errors in the soil uptake by methanotroph influence these results?
P10. In the first paragraph, I encourage using absolute numbers from Figure 3 to describe the reduction in bias, such as (from -13.8 to 8.8 ppbv).
P10. L299-301. I'm afraid I have to disagree with saying that TROPOMI is dense, but we cannot fully resolve the source/sink of CH4. It would help if you had more than CH4 to get OH right (such as HCHO and CO constraints), which has nothing to do with the densityTROPOMI XCH4 observations.
P10. In the second paragraph, you should specifically mention what emissions are used for the retrospective simulations.
P11. Is rice cultivation part of the wetland?
P12. Why talk about livestock in the wetland sections?
P12. 380. Where is the South Sudd in the figure?
P13. Can you provide more physical explanations of why these wetland emission models disagree so much? Is it due to their parameterization or the need for more information about water nitrogen content, heat content, depth of wetland, sulfate content, etc.?
P14. The second paragraph: are we so clueless about the wetland anaerobic activity to use a simple correlation analysis? How about the soil nitrogen, water temperature, depth, oxygen content, etc.?
P14. L421. Please add a fraction of the total for each sector.
P15. L444. Are they missing from other top-down emissions too? The bar is usually low for bottom-up emission inventories, especially in developing countries.
P17. L 515. Does correlation explain causation?
P18. L 560. It's not about the density of TROPOMI data but a piece of factual information from XCH4. We need more compounds not denser data.
P16. is this number of available pixels really a lot? Please provide the percentage for a hypothetical situation when clouds were not present.
Editorial comments:
P1. L16. What do you mean by separately resolved?
P1. L20-21. It is vague; does the hydrological adjustment come after or before?
P1. L22. The sentence (Fossil fuel emission...) is awkward.
P1. L23. Many -> several
P2. L39. What do you mean by strong heterogeneity? Spatial or temporal?
P2. L42. inverse -> inversion
P2. Please use +- for a normal range. 18+-1 is shorter and neater. Please apply this to the entire manuscript.
P8. L242. What do you mean by "spatial source uncertainty"?
P7. Eq.2. The i->j is weird; what do you mean?
Figure 4. The panels are too small.
Figure 5 needs to be enlarged. This is the most critical figure, which is hard to see.
Table2. The numbers in the parenthesis are just the deviation in a defined box, not an actual error. Please inform the readers about it.
Citation: https://doi.org/10.5194/egusphere-2022-948-RC2 -
AC3: 'Reply on RC2', Xueying Yu, 17 Jan 2023
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2022/egusphere-2022-948/egusphere-2022-948-AC3-supplement.pdf
-
AC3: 'Reply on RC2', Xueying Yu, 17 Jan 2023
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 819 | 394 | 28 | 1,241 | 73 | 11 | 14 |
- HTML: 819
- PDF: 394
- XML: 28
- Total: 1,241
- Supplement: 73
- BibTeX: 11
- EndNote: 14
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cited
Xueying Yu
Daven K. Henze
Alexander J. Turner
Alba Lorente Delgado
A. Anthony Bloom
Jianxiong Sheng
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3123 KB) - Metadata XML
-
Supplement
(8014 KB) - BibTeX
- EndNote
- Final revised paper