The authors present an interesting and novel topic of study related to the overlapping of methane plumes. The specific comments are addressed into two separate topics here below:

1) Scientific significance. The overlapping of plumes is a real scenario that can occur worldwide. However, it is not understood the impact of these cases at a global level. Thus, it could be a very specific topic that is only applicable in some examples. Moreover, the proposed methodology is rather complex. Is it possible to just quantify the non-overlapped area of the plumes (of e.g. FIg 2 and 8) and apply (considering the caveats) the IME method over them?

2) Results and validation. The results indicate a strong underestimation in the quantified flux rate when applying the separation methodology. The authors argue in L421 that this could be the results of some pixels not being attributed to any source.  Thus, the results indicate that the methodology needs to be reviewed. 

The manuscript "Separating and Quantifying Facility-Level Methane Emissions with Overlapping Plumes for Spaceborne Methane Monitoring" by Y. Pang et al. documents an interesting topic and analysis regarding the quantification of pollutant or greenhouse gas emissions from industrial point sources based on satellite images. Such a study deserves a publication in AMT.

1) However, the text of the manuscript needs a major improvement and clarifications. 

a) There are signs of a lack of proofreading or of rigor in the writing, such as typos (e.g. at lines 39, 257, 273, 317, 351-352, 420, 426…), clumsy formulas (l.33; "given the same emission rate" at l35; "generally" at l.38, "the high nonlinearity of the Gaussian plume model" at line 78, "accurate priors" at l211, l214-215, "the increases of multi-source Gaussian plume model" at l359-360… ), misleading shortcuts (e.g. l41-42, l317-318…). Some of the potential mistakes can be embarrassing: e.g. between mid-line 217 and line 218, I assume that the authors talk about chemistry rather than about diffusion, and I do not really understand what the link can be between "the low concentration of methane in the atmosphere" and the diffusion or the chemistry. The word at line 237 is "topography" rather than "topology" ? etc. 

b) In a general way, the concepts, plans and methods need to be better introduced and discussed.

From line 68, the introduction loses the clear distinction between the plume detection and the subsequent emission quantification when processing the CH4 plume images in two steps, or the distinction between such a two-step approach, and 1-step approaches such as Gaussian plume fitting (when using the emission rate from this fitting), while such distinctions are critical to follow the manuscript correctly. Lines 72-83 (which have been extracted from section 2) should be improved and better merged in the introduction. The abstract hardly manages to characterize the type of method that has been developed and tested in this manuscript: lines 3 to 7 seem to speak about a full quantification method and line 8 seems to introduce IME as a distinct benchmarking quantification method (and the rest of the abstract does not solve for this potential confusion).  

I thank the authors for having added the results of the quantification of the emissions based on the Gaussian plume model fitting in the result section. However, now, this benchmarking needs to be announced in the introduction or early in section 2, this quantification method needs to be detailed in section 2, and the abstract should probably highlight the comparison between this quantification method and IME. This is a significant result, which tend to confirm that the Gaussian plume model inversion does not behave as well (compared to IME) when tackling turbulent plumes at fine resolution than when tackling mesoscale plumes with > 1 km resolution images. 

c) see the detailed comments below, which provide other illustrations of the general need to improve the text, its quality and its clarity, which applies to all sections.

2) Section 4 makes an effort to synthesize the results from section 3, to provide some explanations for the behaviors of the methods as a function of the "experimental parameters". However, it could be extended and strengthened to better characterize the problematic and successful cases depending on the methods and to provide more interpretation. Furthermore, some items are probably missing such as a discussion on the impact of using the LBPM, and a discussion on the typical range of accuracy of the emissions estimates that can be expected as a function of the observation conditions. Regarding the LBPM, could this approach be hampered by the retrieval noise, or by the regular transition to the RMS metric in the optimization iterative process ? See also some of the following comments that connect to potential gaps in this discussion section.

 

Detailed comments:

- is not the abstract misleading regarding the assessment of using the LBPM ? the results show that it does not impact much the results, while the abstract seems to say it is successful. Section 3.3 does not really highlight the fact that this impact is relatively small, and section 4 does not discuss it at all. 

- the abstract could mention the test on real data which is significant.

- l11: MAPE has not been defined yet; you should rather speak about the average relative error in the estimates.

- l63: how do you derive a "median interval distances of potential methane sources in California" ? I’m curious about it since the result (<200m) is surprisingly small (it is probably a matter of definition for such a term)

- l68-70: unclear; you mean Nassar et al. fix the secondary plume to a specific concentration level corresponding to the secondary source rate given by the emission inventory ? 

=> then you should also discuss other papers using multiple Gaussian plume model inversions to tackle multiple sources with observations similar to satellite images, i.e. something similar to the "Multi-source Gaussian plume" method tested in section 3 (e.g, Krings et al., 2011, which is already cited in the manuscript). I think that the introduction is a bit misleading regarding this: using multiple Gaussian plume model fitting to handle multiple sources is not a novelty. However, by developing an alternative approach with a multi-objective heuristic optimization for the Gaussian plume fitting combined with IME (which is often assumed to behave better than Gaussian plume model inversions when tackling fine resolution images of turbulent plumes) for the quantification, the authors bring new ideas and insights to improve the process of overlapping plumes.

- l96; what does "the improvements of the separation algorithm in missed detection" mean ?

- l124: what could be the difference between sigma_x and sigma_y when using Pasquill stability classes to set-up such parameters? where could sigma_x (and where does sigma_x) stand in eq (2) ? furthermore, the formula for the function sigma_y(x) needs to be given or explained

- equations 3 and 4: we should have C(x’_n, y’_n) rather than C_n(x’, y’)

- l146 The specific Gaussian plume fitting algorithm used here is questionable. Adjusting both u and Q to fit the observations can be problematic since these 2 parameters impact the plume amplitude in a similar way (there is no source of information to discriminate them in the fitting process). Furthermore, the lack of adjustment of sigma_y in the plume fitting may limit the skill of the approach.

- l170-l175 are very difficult to understand

- l281: the question is not whether these sources can be quantified, but rather whether their plumes can perturb the quantification of other sources (?)

- l282: I do not really understand this sentence, and in particular the term "aggressive". It seems to connect to the assumption that the emissions are constant, which makes the quantification problem and the process of the image easier.

- there is still a lack of explicit introduction to the fact that the whole study relies on scenes driven by homogeneous winds, except when tackling EMIT observations. It’s implicitly guessed from the use of theta rather than theta_n in the equations, and by the quick mention to the wind speed at lines 254 (with the clumsy formula: "unified wind") and 285. There is also a lack of discussion on the fact that the use of homogeneous winds to generate the pseudo observations artificially inflates the skill of Gaussian plume models to support the plume separation: there is a piece of sentence about it at lines 461-462 but it is not very clear and it seems to be associated to very specific observation cases.

- it would be useful to show images with different levels of observation noise, in suppl. material if not in main text (the presentation of fig 2 would be misleading if it does not include such a noise: does it ?) to provide an idea of the challenge associated to the plume separation when using noisy images. Actually, one would expect charts with the APE as a function of the level of noise since this level could be one of the drivers of the relative success of SEP vs. UNSEP. In a more general way, the retrieval uncertainty is a critical topic for the processing of plume images and the manuscript should bring more insights about it. In particular, it should provide the values of the level of noise in a more visible way than at lines 266 and 272, and with some justification for such values. What is the value of the retrieval uncertainty in EXP-3 ? Is the % applied to the CH4 background + plume signal? If yes, what is the corresponding background value ? If not, why (note that the full EMIT image is noisy in fig 8), and would not the values 1 to 3% be very low ? What are the typical relative uncertainties associated to EMIT observations ? Could the observation noise explain the relative failure of the use of the LBPM metric ?

- lines 290-300 poorly fit in section 2.2.3; having separate sections dedicated to the detection / quantification method(s) and to the data made available for the detection / quantification would make the presentation of the study much clearer.

- l294-296: this derivation of the wind driving the plume from the wind at 10 m height does not really make sense for the general process of satellite images. This is likely inherited from studies focusing on specific sites where dedicated local measurement of the wind at 10 m are available. In the general case, the wind should be derived from other source of knowledge (typically meteorological analysis) with a better vertical coverage (but less precision).

- l297-300: clarify how it is combined with the SEP approach

- there is no discussion on the CH4 background mixing ratio fields (from sources outside the images, or from small point sources and diffuse area sources within the images), on how it is dealt with in the derivation of the images or when processing the images. Such a background is set to 0 in EXP 1 to EXP 3. How could it impact the theoretical results from these 3 experiments ? Do we see some residual pattern of background variations in the EMIT "enhancement data" ? could it explain part of the biases seen in section 3.4 ?

- isn’t line 322 at odd with equation 14 ?

- line 328: I do not understand why "the quantification of methane source is considered as solving a regression problem". Is not the target of such a quantification the most precise estimate for a given source at a given time ?

- l367: so far, the metric for interference should be the OImass, not "the interference" 

- l393: Fig 6 rather gives the feeling that UNSEP tends to overestimate the sources ? you meant SEP ? what do the person’s R and p values correspond to in this line ?

- I do not understand the correction of the SEP estimations in section 3.4 (l393-395, legend of fig 6): what is the rationale, what is done in practice ? Because of this correction, it is difficult to check whether SEP behaves better than UNSEP in section 3.4.

- l413 vs l422: does the detection and "connectivity verification" underlying the application of the IME method really encompass the full extent of the whole set of plumes ? l413 goes too fast so it is difficult to understand what it corresponds to.

- l447-448: I do not understand the point which is made here, while I believe that the comparison between the Gaussian plume model inversion and IME for the emission quantification as a function of the spatial resolution and scale of the source quantification problem is an important topic

- l450 I do not understand "statistically correct", and the link made at lines 450-451 between the precision of the emission quantification using Gaussian plume fitting and the convolution kernel 

- I do not understand line 471