# General comments

This is a well-structured, well-contextualised piece of work that has the potential to be a valuable addition to the literature. Well done! It focuses on an application of the AirCore concept with custom-built equipment. In test surveys on a coking plant, CH4 emissions of several orders of magnitude larger than inventory estimates are reported.

However, the paper needs work in order to make it reproducible and needs clarification about the use and modification of algorithms and reporting of uncertainty. In addition, the wind measurements appear to give much more consistent results than most papers I have read on this subject - it would be good to see some detail on how this was achieved.

One important note is that the results of the paper relies on work from Yanrong Yang on windspeed correction - this appears to be groundbreaking work and is "in preparation" but needs to be available to reviewers in order for the paper to pass review according to AMT submission rules - I would be grateful if the authors can upload this.

The section on deconvolution is impressively well-written, and I would like to see the rest of the paper up to that standard. More reference should be made to the supplementary material where appropriate - the material mass balance in supplement section 3 is also well written, and another good benchmark for what should be added to the rest of the paper. However, supplementary material critical to understanding and replication of the paper should be moved to the main body of the text, particularly regarding uncertainty and data processing methods.

# Specific comments

The paper goes into a good deal of depth on deconvolution of smoothed mixing ratios of methane and CO2. However, the main impression I have is that this depth is missing in some other critical areas of the analysis, such as how the TERRA algorithm was applied, how modifications to it were made and how its various assumptions and parameters were defined. There is some detail in the supplementary material but more is needed. More detail on the exact methods of fitting the gaussian model would also be beneficial. If someone else was to replicate this paper, they should have all of the detail needed in order to do so. I was also curious as to why TERRA was chosen as to my knowledge it is not widely used in UAV emissions measurements 

Regarding the title - I wasn't clear about what in the measurement methodology is actually new given that the AIrcore concept is well-demonstrated; I would suggest "Development of a continuous UAV-mounted air sampler and application to the quantification of CO2 and CH4 emissions from a major coking plant" or alternatively a better explanation of what is new in the methodology. Even if the methods are well-established, the paper is a good one and a helpful aid for applied science.

It is suggested that methane emissions measured are anomalous - by several orders of magnitude - compared to emissions factor-based inventory reporting. This is a plausible, interesting and important result, if true. However, it seems that this may - and please make this more clear if I have misunderstood - be based on only one measurement flight. Measurement uncertainty is also presented in a way that could lead the reader to assume that it gives an idea of the error  (i.e. accuracy) rather than the precision of an individual measurement.

Although there are thousands of mixing ratios measured and various other components to measurement uncertainty (e.g. uncertainty around wind speed, and hopefully around wind direction although this is not clear), the basic unit of measurement is the flight. Random error will be potential very high when doing atmospheric measurements of this kind - plumes move and change shape in ways that are difficult to account for even within one measurement. Averaging of multiple flights and their uncertainties will allow for quantification of this error.

If there was only one flight conducted in this campaign, then it represents a pilot study, and will need replication and quantification in order to draw conclusions about emissions inventories. Unfortunately it is difficult to make conclusions about any one measurement when we have little idea about the spread. If there were more, then this should be better explained in the text.

In general I would like to see a much more robust approach to quantifying uncertainty and confidence intervals.

# Line-by-line

Line 57-59 - "could" implies that UAVs are not current used - they are widely deployed in commercial operations to quantify methane emissions flux already. This goes back to the question over the title.

Line 141 - Post-processing here is quite casually stated. I would like to see some reference to the detail of Yang's approach here - this itself is innovative and worth going into more detail, but it's also important to clearly describe the methods even if the full detail is elsewhere. The method relies fairly heavily on this correction and AMT requires "in preparation" manuscripts to be available to reviewers, so I think we need to see this paper to be able to assess this part of the method. Windspeed measurement on drones is tricky to get right, and I think the method has to be made more explicit to give the results credence. 

Line 108 /113- the CO2 marker - I was unclear on why this is necessary - is it not possible to map sensed molar ratios to GPS locations without this marker? I assume it's an important part of your technique but I'd like to know more about why

Line 112 - are there disadvantages to waiting longer, e.g. if flights were longer? Is there a difference between samples at the start of the flight that are e.g. 30 minutes old, vs those at the end of the flight?

Line 150 - it is quite uncommon for vertical precision to be better than horizontal for GPS - are you sure this is correct?

Line 155 - can you clarify the discussion of why laminar flow would lead to smoothing of concentration changes?

Line 275 - approximately how long was each flight? Do you expect the plume to have systematically moved during flight, and how do you account for this?

Line 294 - please provide confidence intervals or similar for windspeed and direction. Did these vary at all according to the direction of travel of the drone? Measuring windspeed on drones is notoriously difficult and you appear to have extremely consistent results compared to most other datasets I have seen. Reviewers would need to see the paper in preparation that these results rely on.

Line 300 - how did you handle negative deconvolved measurements, such as those visible on the CO2 graph?

Line 306 - Can you be specific about the TERRA algorithm, your modifications to the algorithm and how it is applied? Kriging of gases in air is a complex business and the use of various kriging parameters, semivariogram optimisation, anisotropic search etc. needs to be clearly stated and reasons given for each parameter choice. I see some information in the supplement, but not enough to replicate your method I think.

Line 308 - The distance between horizontal lines is 15m, yet the kriging nodes are 2m in the vertical. How did you decide on this resolution? I see in the supplement - please refer to it more often! - that there was a 20x reduction from aircraft resolution. Why this magnitude and not more/less?

Line 315 - can you be specific and detailed about how uncertainty was obtained here - I see in the supplement (please refer to it) how you combine them but not how they were actually calculated except through a reference to Gordon et al.

Was wind direction not an uncertainty component? This would likely be non-symmetrical too, as it varies by the cosine of the angle between the wind and the measurement plane. How does TERRA deal with the case where there are two walls that intersect the plume? Is there an implicit assumption that all points in the measurement plane are equidistant from the source(s)? If not, how is that accounted for?

Line 323 - how were delta-mixing ratios arrived at, i.e. what was your baselining procedure? 

Line 323 - It seems plausible given Figure 5 that the plume moved during measurements, and that you may have double-counted it. How is this reflected in your results? Are you confident that there was no double counting? Would this method work on a day that had less consistent wind direction over the 30 minute measurement period? Were there any trends in windspeed duriing the measurement period and how was this controlled for?

Line 330 - does this not apply to the time-averaged gaussian plume, rather than the instantaneous plume that you measured, which should be somewhat non-gaussian or at least with a smaller sigma and higher maximum concentration? How are you sure that you correctly paramaterised the maximum, i.e. found the highest concentration value? Were there differences between sigma y and sigma z?

Line 341 - how was this confidence interval arrived at. Are you using normally distributed uncertainties or something else?

Line 344 - I would like to see TERRA measurement uncertainty reported in the main text by its components and how they were arrived at (normally distributed uncertainty, potential range analysis etc.)

Line 345 - I would not go so far as to say that two models agreeing with each other means that either one is reliable - that's what your validation should do.

Line 358 - was the material balance analysis from daily process data? Where did it come from? please refer to the supplementary material as appropriate - I see there is a detailed section there. Was the production rate constant? Any process variations during the period of measurement? Your measurement took quite a long time I think - do you have any comment on what the maximum length of time for a suitable measurement, given there can be process variations within its timeframe?

Line 372 - please refer to the supplement here so I know to look there for plant information - I'm not entirely clear what taps leakage or door leakage would be even after reading it though

Supplement - Line 69 - These deltas appear to be based on the Monte Carlo from Gordon et al. but those were specific to their  own measurements, and probably quite dependent on the number of measured points. To illustrate - if you have few signal points, wind speed variation can have a big impact, but in big diffuse plumes like in Gordon et al it will make much less difference. You should consider your own Monte Carlo, or alternatively use normally-distributed uncertainties and accept their more conservative results.

I would really like to see more detail on uncertainty propagation here and in the main text. WInd direction is not included here - this can be a large source of potential systematic bias when a plane does not intersect a plume at right angles, and thus biases measurements high according to a cosine relationship. It doesn't seem like Gordon et al go into this either, to be fair.

You use delta uncertainties, which is fine, but these can be nebulous unless well-defined - I would like to understand each one and where it comes from, so that the paper is replicable. The relationship of some of these to the calculation are not clear to me (e.g. box-top mixing ratio, box-top height etc.)

Supplement - Line 77 - This seems highly surprising to me, regarding wind speed. A very accurate and precise anemometer might have a 1% 1 sigma error but surely you are averaging over the entire set of measurements for windspeed? Or does each concentration measurement combine with paired windspeed to arrive at a flux? If that is true, it implies significant assumptions about dispersion of the gas between the source and the point of measurement and these need clearly stated. For example - do you assume that a measurement taken in a gust has correspondingly lower mixing ratios of the target gas, and thus a similar flux to the same situation at a lower windspeed with correspondingly higher mixing ratios?

Supplement - Line 84 - the extrapolation comparison is a good idea. I'd like to see the results reported. It probably doesn't make much difference in your case.

# Technical corrections

Line 95 Supplement - "TERRA" 

Line 101 - these should be given at least one more signifcant figure in my opinion, and differentiation should be made between measured and estimated values

Line 324 - "satellite"

# Last word

I really enjoyed reading this paper, and what I have read is of high quality, and a really interesting and unusual application of methods that I haven't seen before in this combination. It will make a great paper - I hope these comments are helpful and not disheartening. Great work!

Review of “Application of a new UAV measurement methodology to the quantification of CO₂ and CH₄ emissions from a major coking plant” by Tianran Han et al.

This paper presents a new UAV-based system to measure CO₂ and CH₄ emissions from a local anthropogenic source.  I think the paper addresses a relevant question and the presented UAV-based system is a novel combination of methods for quantifying GHG emissions. The paper thoroughly presents the measurement setup and data analysis methods including uncertainty estimation, and compares results from a test flight to different other methods. However, I did not find all the information where I expected it to be and I think the manuscript needs some revisions. I outline my general and specific comments below.

General comments

In many parts of the manuscript, I was missing additional information that I later found in the Supporting Information (SI). I think it would be helpful for the reader to have some of this information in the main manuscript or at least refer to the SI. This applies to some of my specific comments below.

Three pages of the manuscript (out of 15) cover the deconvolution of the measured mole fraction time series. I wonder if dedicating this amount of space is justified. Is the data deconvolution novel compared to previous UAV-based GHG measurements? In this case, its importance could be highlighted. Otherwise, which advantages does the deconvolution method offer compared to a simple response time correction? In contrast, the manuscript contains very little information about the other uncertainty sources and the actual emission quantification method. 

Speaking about the uncertainty sources, these are nicely listed in the SI, table S1. The deconvolution contributes only 1%, and there are other factors contributing at higher percentages. This could be discussed in the manuscript. Also, I was wondering if it would be feasible to measure additional flight legs at the top of the measurement volume box to reduce this uncertainty factor. (Would the wind speed measurements and/or the limited measurement time allow those measurements?)  

I think it would be helpful to refine the structure of the manuscript. The methods section should contain information on how emissions are derived from the air sampler and wind measurements (instead of in paragraph l. 277ff). Novelties or improvements to established methods could be emphasized. Sections 3 and 4 could be merged since they both refer to the laboratory tests and subsequent corrections.

Specific comments

l.14: The first sentence and the subsequent lines in the abstract focus on measuring GHG, it might be good to emphasize the emission quantification right at the beginning.

l.16: The 150m length information at first sounds confusing when mentioned together with the UAV. Maybe add “coiled”.

l.18 “During flights, the air sampler starts sampling as soon as the UAV takes off, and stops sampling after landing.” Is this relevant to the abstract?

l.75/76: Even though using different CO₂ measurement methods, CO₂ emissions have been quantified with a UAV-based system: https://doi.org/10.5194/amt-14-153-2021. Other studies describe measuring mere CO₂ concentrations with UAVs (https://amt.copernicus.org/articles/15/4431/2022/, https://www.mdpi.com/2073-4433/10/9/487, https://amt.copernicus.org/articles/16/513/2023/). Maybe those studies could be discussed?

l.78/79: “three-dimensional measurements of CO₂ and CH₄” To me that sounds like a continuous space-time measurement. “on a trajectory in the three-dimensional space” might be more suitable.

l.88ff: Trying to research the Tier 1 method, I could find neither the “IPCC 2006” nor the “Ministry of Ecology and Environment of of China, 2018” citation in the reference list, but a “Kopp et al. 2021” reference (Physical Science basis of the IPCC 6^th assessment report). It would help to point the reader to the specific report/chapter where the information can be found. Also, it might be worth mentioning that the Tier 1 method is the least complex method for estimating GHG emissions.

l.95ff: It is stated that the new system is based on the AirCore system, which seems to be an established method for airborne CO₂ and CH₄ mole fractions. What has been changed or improved compared to the existing AirCore system?

l.119ff: Which parameters measured by the sonic anemometer are used? Only mean wind speed and direction? Vertical and/or horizontal wind components? Which resolution and why? Is the measured temperature the sonic temperature (as measured by standard sonic anemometers) or the actual air temperature? The text repeatedly refers to “meteorological parameters”, but it is not explained which are used.

l.140ff: I also would like to hear more details about how the correction for rotor-induced air flows and UAV attitude and motion was performed. The Yang 2023 paper is not published, so no details are given about these corrections. Related to that, which instruments were used to determine the UAV airspeed and attitude, the GPS sensor (which?) mentioned in the section after?

l.160 ff, Fig. 2, Fig. 3: It would be helpful to be more consistent with how you name different instruments/setups. Using different words to describe the same thing makes it hard to understand the validation. (E.g. Picarro=online measurement= CRDS =”same analyzer”?, air sampler = presented system?, Artificial source = lab air = mix of standards of CO₂, CH₄?). Did I understand correctly that adding the Picarro in Fig2a is the validation reference, while the general process in Fig 2a and Fig. 2b shows the standard process of the air sampler? I also did not understand the purpose of the zero air in Fig 2b (it is mentioned in l. 197 for the first time).

l.175: I think “good agreement” could be supported by a correlation plot or quantifying a correlation coefficient and errors. These numbers might also show if/how the correlation improves for the convoluted signal introduced later.

Fig. 3: Also related to the naming of the signals, it could be helpful to mark which lines belong to which original signal and be consistent with assigning labels. Additionally, it might help to include a correlation plot instead of the current Fig. 3b and make Fig. 3b a separate figure.

l.183ff: Is the deconvolution a novelty added to the original AirCore sampler? In the section above, the Karion et al., 2010 study is cited listing different error sources. Were the error sources quantified and/or mitigated in the original AirCore methodology?

Fig. 3b: Please label the axes more clearly (unit of time on the x-axis, normalized concentration on the y-axis?). The two plots could also be merged into one with two curves, where the colors match Fig. 3a.

l.274: It seems like the elevation of the coking plant is nearly sea level, but it might make more sense to indicate flight altitude in AGL.

l.277: This paragraph (maybe better located in the methods section) needs further explanation. Which input parameters are used (specify “meteorological data”)? The control volume is defined by the flight path of the UAV? Are all terms in the mass balance (Table 2 in Gordon et al., 2015) considered? Why does the UAV not measure in the plane of the top of the box (which is later on listed as an uncertainty)? In the SI, it is explained that the vertical transport through the top of the box is neglected, but it might become important under unstable conditions, so information about atmospheric stability during the measurements would be interesting.

l.284: How is the original TERRA further modified and tailored?

l.297: What are the CO₂ markers? They are shown in Fig. 4, not Fig.5.

l.321: “The uncertainties for the estimates were derived from detailed analyses of each uncertainty source”: Here, I would like to hear more details about how the different uncertainty sources contribute to the overall uncertainty.

l.387 – 389: The last sentence might be a bit too general since only one measurement case is studied. Also, your results show that the IPCC emission factor was too high only for CO₂, not for CH₄.

Technical corrections

I could not find some citations in the reference list (e.g., IPCC 2006; Ministry of Ecology and Environment of of China, 2018; European IPPC Bureau, 2001). Please make sure that the reference list is complete.

Please make sure to spell proper names correctly (AirCore, Gålfalk, TriSonica).

The authors might consider using “uncrewed aerial vehicle” instead of “unmanned aerial vehicle” if this is not a firm convention.

It might be helpful to use consistent wording, e.g. do “mole fraction” and “mixing ratio“ mean the same? Is “air sampler” the same as “Aircoil”?

l.21: Please spell out CRDS when used first in the abstract.

Fig. 1: Pump instead of bump?

l.123: Is Geotech the manufacturer or just a distributor?

l.168: Please spell out MFC.

l.292: Deconvolution is explained in Section 4, not 3 (if sections remain unchanged).