A comparison is presented of three methane (CH4) retrieval products for the TROPOMI instrument. Inversions are performed using the TROPOMI data, the CHIMERE model and a variational data assimilation system. Substantial differences in European fluxes were derived for the inversions using the three TROPOMI retrievals (SRON, BLENDED, WFMD) and these each differed from inversions using the surface data. This result has important implications for the use of TROPOMI data in regional methane inverse modelling.

Overall, I find the study to be important, timely and thorough. However, I feel that the structure of the paper could be improved for clarity and brevity. In particular, I would urge the authors to consider the following:

1. Are all OSSEs strictly necessary? In particular, I wondered if the “diff” OSSEs could be cut (or at least moved to the Appendix/Supplement), without detriment to the paper.
2. In many places, the structure and text could be improved for readability (see suggestions below).

General comments:

1. I don’t understand why the WFMD product leads to substantially lower fluxes than the other inversions. The terrestrial mole fractions seem to be between those of SRON and BLENDED (Figure 9), and the lateral boundary conditions seem to be similar to BLENDED (D5). The authors address this on L532: “Therefore, the strong negative increments on the Inv-WFMD fluxes result of a complex balance between the local gradients of the increments on the background and on the fluxes: the system could have difficulty separating both when using the WFMD observations.” But can it really be that complicated? If the boundary conditions are roughly similar between two products, but the terrestrial mole fractions are mostly higher for one (Figure 9), then surely, the fluxes for that product must be higher, not lower? Furthermore, it’s not clear why their explanation (that the fluxes and background can’t be easily separated) would only apply to WFMD. I wonder if there could be a bug here…
2. Throughout, why are posterior flux uncertainties not provided, except for in Figure 9 monthly means?

Specific comments:

Many statements in the abstract are ill defined, or vague:

L 7-8: it’s not stated what these increases or decreases are relative to

L 8: “Seasonal emissions are highly correlated across the inversions.” I’m not really sure what the point of this sentence is, or what it means. Cut?

L10: What does it mean that the boundary conditions differ “substantially” for WFMD? Also, is this true? It doesn’t seem so from Figure D5.

L10: “Evaluation with independent surface stations shows error reduction for about half of the sites, with BLENDED performing best”. I think this means that the residual between the observations and the posterior mole fractions is reduced for about 50% of the monitoring sites. But BLENDED showed a reduction in residual for more sites than the other inversions? More precise language is needed.

L11: “However, no product is systematically closer to the emissions estimated when assimilating surface observations”. This is too subjective a statement. In any case, to me, the SRON and BLENDED inversions looked very similar to the surface inversion for ~10 months of the year, whereas WFMD differs from the surface inversion most of the time.

L13 and L14: What “errors” and “quality filters” are being referred to here?

Main text

L22: “with higher rates OF INCREASE over the…”

L39: “relative”, rather than “relatively”

L45: Shouldn’t a range of wavelengths be provided (“spectral range”)

L124: Briefly (1 or 2 lines) describe the de-striping procedure

L137: We keep only THE highest quality…

Equation 1: define sigma and sigma_hat

L164: Couldn’t it be confusing to use Δx here? It could imply only in one horizontal coordinate. Just say within 0.01 degree lat/lon?

L180: This is the first reference to a figure (Figure 5). I presume the journal will require that figures are referenced in order?

L208: Why is this interesting? It doesn’t actually say, but seems to be implying something. Is this sentence needed?

L215: “e.g. in Scandinavia”. Be more specific: what are the patterns in this region.

L216 “The temporal variations… show consistent patterns across the products and align rather well with GOSAT”. What is the basis of this statement? To me, GOSAT looks very different to SRON and WFMD, but similar to BLENDED.

L234: Is this really an order of magnitude? Isn’t it about a factor of 3?

L240: I don’t understand how the 0-5% difference is quantified to a profile. Is this per level?

L325: Do you really mean “surface roughness” here? If so, you could use surface roughness (the meteorological term) as a filter, rather than topography…

L390: If the dataset is split randomly, isn’t there going to be substantial correlation between the testing and training sets that influences the metrics? Most of the testing set will be adjacent to points that have been used in the training. It would be preferable to have the testing and training set be separated in space/time (or some other factor).

Figure 7 caption: this figure needs explaining more thoroughly

L413: since this is the first line of a paragraph, restate this initial sentence so that it reminds the reader what you’re talking about.

L450 and elsewhere: try to avoid subjective terms like “best”.

L472: You can’t say that the inversion “correctly” fits the data as there will always be errors. Just say that the residual is reduced after the inversion (which it must be, if the inversion is working correctly).

Figure 10: why is this the only place where emissions uncertainties are provided?

L506: “Yet, the amplitude of the emission peak raises questions about its origin.” This sentence implies something but doesn’t spell it out. What are the questions, what could be the origin? Or do you just mean that you don’t believe that this peak is real? If so, say so, and justify your reasoning.

L507, L512 and elsewhere, don’t start paragraphs with “However,”, “Moreover”, etc. Each paragraph should tackle a single idea. These words imply a continuation of an idea.

L556: Avoid “better”

L602: “It can appear in the optimization process in variational inversions.” This is a sentence fragment. Reword.

L606 – 629: Could this section be removed?

L635 – 637: I don’t know what this sentence means. Rewording is needed for clarity.

Conclusions: I think the conclusions are far too long and contain several statements that aren’t really justified, given the results. The authors should refocus this section for concision.

L661 – 662: This one-sentence paragraph doesn’t seem necessary for a conclusions section. It doesn’t really say anything.

L675-676: What is meant by a “proper formula”? Wouldn’t it be better to say what is physically needed here. I.e., how should the error be appropriately calculated.

L678: Are model biases really a key outcome of this study? Do we need to state this here?

L679: “Refinements in the configuration of the inversion system are thus also essential to enhance the consistency and robustness of emission estimates.” What does this mean? Why is it essential? What about your results indicates this to be true? (If you are just saying that inverse modelling systems need to be improved, you can cut this, as it’s well understood, and you don’t address this in your paper).

L693 – 694: Your results don’t show why we need higher resolution, and, clearly, global inversions would be preferable. You can safely cut these lines.

L697: Does your work really suggest that joint in situ and TROPOMI inversions would improve matters? At the moment, you show major differences that are difficult to reconcile. Perhaps, if we knew how to appropriately reconcile these systematic differences. But at the moment, I wonder if your work shows that in fact, we’re not ready yet?

Supplementary material

Figure D4: Is this missing a 4^th row (mentioned in the caption)?

Review of the article titled: Can we obtain consistent estimates of the emissions in Europe from three different CH4 TROPOMI products?

General remarks:

At the beginning I would like to underline that this article is definitely worth publishing as its message is of the most importance while satellite products like CH4 concentration distribution are becoming more and more popular and conclusions taken from those measurements have political and economic importance. Authors did extremely good job and provided variety of analyses regarding three basic products related to CH4 column concentrations. The main aim of this article is to answer the title question. The role of scientific article is to give evidences and logical arguments with statistical approach to set up the thesis and confirm it (or discard it).

The scientific level of this article is very high. Large part of the paper describes methodology applied to obtain particular products from TROPOMI instrument. Secondly, modelling of the methane fluxes was described. This was done to bring the final product, i.e. CH4 fluxes to the readers.  From the point of view of scientist not working with satellites measurements the paper is very long and frequently refers to other publication describing in detail particular techniques and methodologies. In my opinion it resulted with a patchwork of information not deep enough to understand the products without reading those referred publications and making the storyline heavily disturbed.  All this looks like authors wrote the paper and gave the title question at the end. I bet that readers of this article will be curious what is the answer to the question. And presentation of that problem should be a main goal of this paper. Or authors can change the title easily to definitely more scientific oriented, like “comparison of three products…..”, where such structure of article will be consistent with the title.

The consistence of flux/emission  product should be thoroughly analysed with detailed prepared tools. This is where OSSEs helps a lot and article clearly benefits from this approach.  Consistence is also expressed in statistical way, and here the maps are extremely helpful.  However to answer the question authors should present a clear and trace line leading from methodology to consistency at pixel, region, country and continental scale.  Here, analysis is dissolved in many technical references which are not relevant to the final answer.  

I see clear lack of hypothesis verification and many subjective statements not explaining if we can or cant obtain clear consistency of 3 TROPOMI products.  My expectation for this article would be a detailed discussion what affects consistency of the 3 outputs  in terms of methane flux distribution (temporal and special).  

• Discussion of similarities and differences between the products. Those only which are relevant to final conclusions.
• Discussion of metrics applied to judge consistency of 3 products or any pair of those.
• Discussion of results in form of the prove of the hypothesis with detailed description of the discrepancies and similarities of the product results (final ch4 fluxes)
• Discussion of possible errors and uncertainties of the fluxes and its relation to consistency.
• If authors find it necessary comparison with ground-based observations which in my opinion is irrelevant to the question (in title).

One would expect that the test of consistency can be done using known emission distributions for methane (large point source and large area sources). There are some  verified sources of methane which can be potentially used for such purpose. However, again, agreement with real data is not matter of product consistency but rather reliability. Authors should look at the comparison of the products and condition when they deviate from consistency. That would be an answer to the title.  In the form they proceeded it’s a mess.

Specific comments:

Abstract.

Abstract should be a kind of teaser for what reader will find inside the article. It is. However, the expectation of answering the question is not confirmed in it. One can have feeling that the article will contain a chaos of information what indeed is true.

The obvious comment that XCH4 differences are related to aerosol and albedo one can find in any of publication related to any of those products. Authors should be a bit more constructive here giving something new in that field (e.g. quantification of those effects).  

Comprehensive abstract should not contain contradictory statements : e.g. 

“European CH4 emission budgets show relative increase of 2%, -1%, 33%...”  - whatever relative increase is, its clear products are not consistent.  Or they are, because 33% in compare to 2% is not much?

“Seasonal emissions are highly correlated…” – so, probably we will look for bias in products. So, if the correlation is a measure of consistency the answer is yes, if bias – the answer is no.

“Using consistent error definition….for more consistent emission estimates” – what is consistent error definition at this stage its hard to understand but it suggest that it will play a key role in the answer. But it is only “….paving the way…” so most probably the answer is again: No.

Article will be much more understandable if authors will focus on main message which is the answer to the title question and will build the well structured hypothesis verification.

1)Introduction.

First few sentences (lines 17 – 24) are so much obvious and frequently repeated. Its now primary school level knowledge.  I think this article is not giving any new concept about those topics and as being submitted to ACP, every person who will read it will be aware of those basic facts about methane.  

  Next few sentences (lines 25 – 29) are referring to BU, which is later not compared with nor used in any form. Authors also do not provide any input to this method of methane budgeting. Critical approach to the BU EU methane budget is not relevant if there is no better proposition from authors side.

Definitely there is a lack of introduction about satellites, especially those used for XCH4 measurements. There is a whole constellation of such platforms nowadays and role of S5P in that landscape would make a good background for this paper. Please, describe or list all products currently used for XCH4 retrieval. This will help to understand the problem of coherence of product.

Secondly, please discuss where the products were applied (examples of application of those products – like lines 40 and 41, but with deeper explanation of problem in term of coherence of those products).  How much those products differs for O&G facilities, mines, landfills…

Be specific while mentioning some aspects of work:

Line 48 – prior vertical profile of what parameters?

Line 49 – stratospheric background of what?, aeroslos… - do not leave “…” here. Discussion of those effects are essential for this paper and should be introduced carefully in that chapter (instead of GHG effect).

Line 49 – what retrieval?

Line 50 – scattering of what? What spectral range? What issues are mentioned here?

Line 52 - …routinely updated. – provide the reference.

Line 54 – avoid slang, like “beta research product”.

Line 54 – At this stage destriping algorithm might be at least briefly explained.

Line 67 - …the consistency and applicability… those terms require the measure (scale) would be nice to have few words in introduction about it.

Please explain why Europe is a targeted area for this article and why 2019 was chosen. It is important because later authors discuss a seasonal effect, what in relation to single year is not usual. Especially that there is a definitely longer period available now (2025). Effects in that year might not be representative for a “usual” seasonal singals.

2) Data and methods

Here, my opinion is that authors can easily shorten this long chapter giving a list of good references to each product.  Only aspects important for a latter discussions should remain.

Lines 100 – 108 – are those details important for the products coherence?

Line 115 – This regularisation is more frequent referred as Tikhonov, or Tikhonov-Philips.

 Again level of description is strange. We cant see any reference to this algorithm anywhere later.

Line 116 - …state vector… not explained what it is and how it may affect coherence?

Line 119 – VIIRS instrument is certainly important but does use of it influence the coherence of the products?

Line 123 – …(qa_value… later in line 137, and variable xch4…in line 152)  is such variable need to be mentioned here? If yes please describe what it is.

Line 124 – why do we need to know that there is 4731034 observations – how this number affects the conclusions?  Table 1. contains this information.

Line 124 – if destriping is applied to only “new” data and it is important for coherence in 2019 – why authors didn’t insist to apply it also for 2019?

Line 126 – what is the single sounding precision?

My comment to authors regarding uncertainty calculations: It is a crucial part of this paper and authors underline it the conclusions. Why it is so poorly explained in the chapter 2. Please devote some space to explain directly how you apply the uncertainty chain in your calculations. Give a table with some values.

Line 133 – TCCON observations bias regarding GOSAT not TROPOMI product is not referenced. Explain why do we need this correction and how does it affects the consistency of products?

Wouldn’t be good if TCCON is used for product consistency check?

Line  136 – limitations are extremely important for this discussion – no references, no comments?

Line 137 – qa-value – what product it refers to ?  SRON?

Line 137  - …recommended by the authors.  what is rationale behind that?   Authors of what?

Line 141 - …provided errors… please give a reference and describe more specifically.

Line 158 – Section 2.1.3. contains very poor description referring partly to section 2.1.1.! And defines as st dev of retrieval noise – not explaining what it is and how does it is calculated. In terms of XCH4 values. Please, provide this information somewhere.  Also pseudo noise should be defined if it is relevant to final conclusions.

Equation 1 – no variable symbol definition is provided.

Line 169 - …is relatively mature product, even though sparse. - If it is important for BLENDED product please indicate whet does it mean.

Line 174 – 9.2 ppb bias is repeated ?

Line 194 – what scene description variables are important here?

Line 195 – Spatial patterns of differences…. - Of what?

Line 199 - …change of pixel size. – why this  year was chosen not e.g 2020?

Line 203 – If CFC was exceptional, maybe longer period of comparison (2019 – 2024) would be beneficial to cancel such effect.

Line 205 – where is a Figure S1?

Line 206 – there are statistical tests for non parametric hypotheses?  What does “approximately” mean here?

Line 209 – 9.2 ppb mention 3^rd time

Line 210 – why there is no such comparison at the Figure?  Its not clear how the GOSAT average is calculated if there is so big difference between the pixel size and number of data?

Line 213 – There are over statistics describing distributions not only average (expected value).

Line 220 – how does offset is calculated (difference of averaged monthly values?), why not calculate it for each products pair.

Line 220 – what does approximately means  - why not to give st dev?

Line 230 – what noise authors refer to?

Line 232 – what is “pseudo noise” ?

Line 234 – 3.9 is not order of magnitude lower than 12.2

Line 234 – scaling up the “error” is unusual procedure that requires additional clear evaluation – please provide the reasons behind this transformation. Being smaller is not one.

Line 235 – Author should explain why Eq1 is applied and where does those coefficients come from.

Line 236 – no definition of error is given (anywhere), how can be error understood here : uncertainty, systematic error (bias),

Line 237 – also here observational error is without definition…please provide.

Line 244 – Huge difference of XCH4 (200ppb) is hardy evaluated by fig 4. Can authors provide a panels with a view on higher pressure levels so this, most important part of profile is clearly visible?

Line  275 – please give a reference  (…less than 2 weeks)

A whole chapter 2.3. should be target at application of products, its importance for emission estimation and coherence of the emission patterns. Details can be linked to literature and equation 2 is not necessary, or authors should refer latter to it discussing the sensitivity of model to pressure profile in particular products if different.

2.4. chapter is again basing on many references and construction of inversion model principle is not relevant if not used later in results chapter.  This article is not basing on particular matrices used in eq.3.  If statement in line 231 can be explained without introduction of the Eq 3 than authors may revise need of that equation and chapter 2.4.1 and large part of 2.4.2.

Line 305 and 306 – why  usually 20 – 30 , what is the cost function , is it relevant for this discussion?

Line 315  - is it RSD? Why set it to 100% ? How does it correspond to “error” definition in model?

Line 318 – why 2% not 5%

Line 319 – What does “large” means in reference to model error?

Line 325 – what is “poor” quantitatively? How does it affect XCH4?

Line 331 – Confirmation is not supported by logic and statistics, model can run wrongly in some areas and this will also result as same distribution of removed observations in all products.

1. Results

I personally found this chapter very interesting and well prepared. It contains a lot of earlier briefly or poorly explained information and references. It looks like it was written by different person or group than chapters 1 and 2.

My only general comment in relation to the results is to focus on coherence of products (like fig.7). Give as much quantitative comparisons and avoid subjective statements like “could be related” , “possibly” etc.

Chapter 3.3  - Seasonal cycle and annual budget are the parameters which are affected by boundary conditions and exceptional situations if they are referred to a single year period. In my opinion authors could make the analyses longer for this comparison.

Line 506 – Figure D5 is appearing before Fig D4 is mentioned (line 520)

Line 612 – There is no information what is uncertainty of calculated yearly emissions and it is hard to judge if change from 25.2 to 25.3 is statistically significant. Definitively term  “slight” is not quantitative.

1. Conclusions

I will repeat my general remark here. I would like to know the answer to the title question. If it hasn't been given in chapter 3, I expected it to appear here.

Line 654 – what does the “holds great promise” refers to – list a reference or be more specific.

Line 658 – Why consistency of products is a crucial for interpretability? Why does one product cant be just better than other two?

Line 660 – if the lower quality of data would be chosen – coverage factor would be greater, isn’t it like that? This both parameters are basically inverse dependent. Please give a measure that change of data quality filter affects the emission distribution. It wasn’t in the result, beside the counterintuitive statement that larger errors results in enhancements of the emissions.

Line 661 – what deeper understanding  refers to?  Please, provide the quantities for each factor in which it impacts the retrieved fluxes:

Quality filtering,

Spatial coverage,

Spatial distribution,

Uncertainty of observation,

Ability to constrain emission (not discussed earlier).

Line 665 – Application of ML to discover that retrievals are dependent on aerosol, albedo and striping is not new – all that was clearly reported in cited literature. Why it is not an object of this investigation? It is what I expected.

Line 666 – Recent…. – why not to calculate how much those activities (developments) can improve the emissions coherence (this would be a great answer to the question).

Line 681 (probably most important one) – Without the uncertainties this numbers are not giving proper answer to title question. Please set the hypothesis and confirm or negate it with proper statistical analysis.