Summary

The authors present measurements of methane emissions from concentrated animal feeding operations (CAFO) with MethaneAIR, an aircraft-based imaging spectrometer and airborne demonstrator of MethaneSAT. They use a "targeted scene-based" approach, defining subsets of MethaneAIR data located around the CAFOs prior to emission quantification to imrpove plume detectability. Wavelet denoising is applied to those targeted scenes to retrieve an higher amount of detected plumes within noisy background, which are afterwards classified as originating from a CAFO or not. Emissions of single agricultural plumes are estimated via flux divergence integrals for a series of "expanding boxes" placed around the source location. Only robust emission estimates (rel. uncertainty < 100%) are used for further analysis. 

Per animal emissions across CAFOs within Colorado are calculated using reported maximum livestock numbers for each CAFO. Dairy CAFOs were found to have higher per animal emissions than beef CAFOs on average. Both estimates by far exceed the bottom-up reportings using corresponding national and international (IPCC) emission factors. Other measurements also report higher emissions than bottom-up, but lower than the authors calculated using MethaneAIR data. The authors investigate, that higher emission rates, and moderate wind speeds for expected low emissions, are favorable contidions for robust emission estimates and further, that high-emission outliers possibly arising from a combination of high wind speeds and site-specific factors, like manure management. The limited representativeness of static emission factors is demonstrated, while more data and a better knowledge about actual on-farm activities are found to be key for improved emission driver studies. Methods used are considered as operational blueprint for quantification of agricultural emissions with similar airborne or spaceborne instruments.

General and major comments

The manuscript treats several important topics, namely the sector-specific quantification of agricultural emissions within a complex emission landscape (surrounding oil and gas facilities), the potential of using remote sensing instruments for quantification of areal sources (multiple, wide-spread (low-emission) CAFOs) - in contrast to the previous focus on (strong) point sources, and the mismatch between top-down and bottom-up emission estimates for certain sectors/regions/situations (here CAFOs in Colorado). The idea of an operational blueprint for similar approaches is considered of being valuable for the scientific community. However, i can only recommend publication after the authors address some major issues listed bellow: 

1. Given that the authors consider their manuscript being an operationl blueprint, the description and evaluation of applied methods is covered only superficially. Also the title "Agricultural methane plume detection using MethaneAIR: A targeted scene-based approach with wavelet denoising and divergence integral methods" raises expectations on detailed discussion on how and to which degree the targeted scene-based approach and the wavelet denoising improves emission qunatification of CAFOs, which are the focus within this work.
   
   However, the targeted scene-based approach is only briefly explained, being a rectangular subset of MethaneAIR L3 data, centered around a known CAFO location with an extent of +-0.02° (~2km) in every direction. Why was exactly this scene size chosen? I guess some strong plumes may have an (detectable) extent reaching far beyond 2km downwind of emission location, others, originating from sources with low emissions or measured during low wind speed conditions may require smaller scene sizes to be precisely determined? A comparison of detected CAFO plumes (and estimated emissions) derived with and without the targeted scene-based approach (using entire mosaic) would be beneficial, e.g. for some case studies. Or at least a qualitative discussion related to that. 
   
   The same holds true for the wavelet denoising: No details on the wavelet decomposition are given. What is the "noise-threshold", which is substracted from the scene (i.e. which amount of CH4 fluctuations are considered background variations / instrument noise, which are considered a plume signal) ? How does the mentioned fliter algorithm mitigate false detections? The authors do not give details on this method, despite the fact that it seems to be new, with only a single reference to Zahng et al. 2026 (preprint under disussion). 
   
   The divergence integral (DI) method from Chulakadabba et al. 2023 is directly adopted to infer emissions, however Chulakadabba et al. 2023 estimated the limit of detection (LOD) for this method to be 500 kg/h. The authors mention that they target sources with estimated emissions of 100 kg/h and less (see Section 2.1).  Hence, the application of DI seems not to be straightforward for the targeted CAFOs. The authots should either clarify, why they have chosen DI despite several CAFOs are expected to be bellow LOD, or demonstrate, that DI also works realiably for emissions as low as 100 kg/h. 
   
   Given that the novelty of the combination of trageted scene-based approach, wavelet denoising and DI is several times mentioned in the manuscript, and seems to be the focus of the work, the above mentioned relatively poor evaluation of the application of those methods should be improved in a revised version, ideally alongside an estimation of an overall LOD for their combined method. This is especially recommended, when considering my second major comment (see bellow).
2. One finding is, that emission etimates for some CAFOs were significantly different when the CAFO was studied under different meteorological conditions or simply during different overflights (Figures 5 and 8). If robust emission estimates are more likely for situations with high emissions scenarios (e.g. high wind speed), than situations with low emissions might simply be falling bellow the (unknown) LOD (not detected as robust plume) and hence biasing CAFO emissions high. This could contribute to the discrepancy between EPA/IPCC, other top-down estimates and the estimate with MethaneAIR for avereaged per animal emissions across Colorado (Table 1). Some discussion on that would improve the robustness of MethaneAIR CAFO emission estimates. This is all the more important, since different reasons for the discrepancies between MethaneAIR and EPA emission estimates (see Table 1) are being discussed, of which some seem to be rather speculative. Especially the statement, that beef CAFOs may operate bellow maximum capacity, whereas dairy CAFOs would not, seems to have no scientific evidence. 
3. The conclusion, that most MethaneAIR observations of beef CAFOs are in line with EPA estimates (line 261-265) is lacking sound reasoning, and contradicts the findings presented in Figure 5 and Table 1. (I count 6 mean beef CAFO  observations being higher than EPA in Figure 5, and only 3 are of same magnitude, same result from Table 1, where MethaneAIR estimates for beef CAFOs are approx. 5 times higher than EPA, still 3 times when using MethaneAIR median). Later in line 284, the authors state that their beef CAFO estimates "remain above EPA values". Those contradictions should be clarified. 
4. From my perspective one additional drawback of the manuscript  is, that no total emissions estimate of an entire CAFO is shown (= per CAFO emission), nor a comparison of total emissions from different CAFOs. Instead, in all cases only per animal emissions are reported. Given my concerns about weak evaluation of applied methods (see first comment), an overview of how much a typical CAFOs as a whole does emit, and a comparison to emission strengths observed in other MethaneAIR studies (e.g. Chulakadabba et al. 2023 or Zhang et al. 2026) would increase credibility of emission estimates. The pure focus on per animal emissions is in my opinion too narrow and targeting solely emission factor/bottom-up evaluation. Since those estimates are based on old (2017) reported values of maximum capacity of animals per CAFO, they are likely missleading to some degree anyway, whereas a per CAFO estimate would not rely on assumptions (emission factors). The authors themselves note, that a more robust knowledge on actual livestock numbers is critical for improved analysis, hence i would all the more suggest, to include "lower hanging fruits" of per CAFO emissions . 

Minor comments

Abbreviations for "oil and gas " and "concentrated animal feeding operations" are used several times within the introduction, but are introduced too late/ at wrong positions. E.g. line 102: "...to quantify O&G point sources...", but explanation of O&G is only given in line 113/114. Similar for CAFO abbreviation and others. 

There could be introduced more studies about aircraft-based estimates of agricultural CH4 emissions. By now, the only one mentioned is McCabe et al. 2023 (line 116). Espcially with regard to evaluation of applied methodologies (see first major comment), a discussion of MethaneAIR retrieved agricultural CH4 emissions in the context of other similar approaches could be interesting for the reader.

Figure 1: Caption lists subplots a), b) and c), however the letters are not assigned to the corresponding subplots. 

Figure 1: Why is O&G inventory shown. If the intention is to illustrate the spatial overlapping of O&G and agricultural emissions, it should be also briefly mentioned in the text.

Line 152: Missing parenthesis at the end of "(~17–20 ppb over flat terrain at 10m x 10m, (Chulakadabba et al., 2023)"

Line 180: Where is your wind information in that processing step coming from ? HRRR?

Line 187: missing words: "the same divergence integral method ... in Chulakadabba et al. (2023) is applied as" --> e.g. "the same divergence integral method as introduced by Chulakadabba et al. (2023) is applied as"

Line 201: "oil-and-gas" <-- written different this time, better "oil and gas" or "O&G" for consistency.

Figure 4: Why are there so many total scenes and plumes detected in Colorado, but not in other states? My impression from Figure 1 would be, that there is an equally high number of CAFOs flown by in Texas, Oklahoma and Kansas. Or were conditions worse for those flights, because of e.g. unfavorable synoptic situations/weather? Additionally, scenes and plumes in Arizona are listed, however flight tracks shown in Figure 1 do only pass through a very small piece of northeastern Arizona, far away from any indicated CAFO. Were there so many small CAFOs only detected during flights?

Line 264: wrote EPA beef estimate to be 6.57, but in Table 1 it is listed with 6.67. 

Line 319 and following: Since the amount of data used to infer ideal wind speed regimes is low, i would suggest to avoid specific numbers (e.g. 2-6 m/s for emission < 100 kg/h) and rahter give a qualitative conclusion: "we found, that CAFO plumes with low expected emissions most likely passed our robust plume criterion for moderate wind speeds, whereas CAFOs with higher expected emissions were detected also for calm and strong winds".

Line 337: Is it really farm 950? In dairy subplot i do not see the proposed elevated emissions for farm 950, and in beef subplot i see it, but for fram ~945. Maybe indicate with small arrow which one is meant. 

Line 352: temp --> temperature

Line 368/369: "that would otherwise be statistically overshadowed by dominant oil and gas emissions." This seems to be speculation, since it was not demonstrated by comparison with e.g. wavelet denoising applied to whole mosaic. See major comment 1 above. 

Line 372: Speculation on operation closer to maximum capacity for dairy CAFOs compared to beef CAFOs. Either treat it as one possible explanation (remove "likely"), or add references or analysis supporting this statement. 

Line 376: you mention that more data is necessary for improved driver studies etc. You have data for other states, though significantly less than for Colorado. Do they tell a similar story? 

 General Comments

This manuscript presents a novel application of MethaneAIR airborne data to detect and quantify methane emissions from concentrated animal feeding operations (CAFOs) across the United States. The authors develop and apply a scene-based extraction method combined with 2D discrete wavelet denoising and a Gaussian-based Divergence Integral (DI) flux quantification approach. The work demonstrates that agricultural plumes, generally weaker than oil and gas signals, can be isolated and quantified by spatially subsetting flight mosaics around known CAFO locations, and that measured per-animal emission rates frequently exceed EPA inventory values, particularly for dairy operations.

The topic is timely and policy-relevant, given ongoing uncertainty in agricultural methane inventories and the increasing deployment of airborne and spaceborne methane sensors. The methodological contribution, combining scene-based subsetting with wavelet denoising to improve detection of low-magnitude agricultural plumes, is a useful extension of existing MethaneAIR workflows. The paper is generally well-written and the figures are informative.

However, I have several major and minor concerns that should be addressed before the manuscript is suitable for publication in ACP. The most significant issues relate to: (1) the quantitative treatment of uncertainty; (2) the use of maximum permitted animal capacity rather than actual animal numbers as the denominator for per-animal emission estimates; (3) insufficient characterization of the background removal procedure; and (4) a number of conclusions and comparisons that appear to be overreaching given the data available. Addressing these points will substantially strengthen the paper.

Specific Comments Major Comments

1. Use of maximum permitted capacity as a proxy for actual animal numbers (critical)

The per-animal emission estimates throughout the manuscript (Figure 5, Table 1, Sections 3 and 4) are computed by dividing observed plume fluxes by the maximum registered animal capacity reported by the Colorado Department of Public Health and Environment (CDPHE, 2017). The authors acknowledge that CAFOs may not operate at full permitted capacity and that the CDPHE data are now seven years out of date. This is a fundamental problem that severely limits the interpretation of the per-animal results. A farm operating at 50% capacity would yield emission estimates twice the EPA factor even if actual per-animal emissions were entirely consistent with inventory values. The authors must either obtain more representative animal count data (e.g., from state livestock census records, permit renewal filings, or satellite-based footprint estimation as used in some recent studies) or, at minimum, perform a systematic sensitivity analysis showing how the per-animal conclusions change across a plausible range of actual-to-permitted capacity ratios. Without this, the statement that measurements "exceed EPA emission factor estimates" for dairy (Section 3, Table 1) cannot be taken at face value. It is equally possible that the exceedance is an artefact of the capacity denominator rather than a genuine emissions signal.

2. Background methane field removal and spatial correlation of residuals

The manuscript does not describe in sufficient detail how the background XCH4 field is removed prior to plume detection and DI flux calculation. For the wavelet denoising step (Section 2.4), it is unclear whether a scene-level background subtraction is applied before or after denoising, or whether the wavelet approach itself implicitly handles the background through low-frequency component removal. This distinction matters because any spatially correlated background signal (e.g., regional methane enhancement from oil and gas fields) will propagate into the DI flux estimate. The authors should provide a dedicated methodological description of their background treatment, including any spatial smoothing kernels or polynomial fitting that is applied, and demonstrate with at least one example scene that the residual background is spatially uncorrelated at the scene scale.

3. Uncertainty quantification and robustness criterion

The robustness criterion, a ratio of the standard deviation to the mean DI-derived flux less than 1 (i.e., relative uncertainty below 100%), is quite permissive for a study making quantitative comparisons with inventory estimates. A plume with a central estimate of 200 kg h⁻¹ and a 1-sigma uncertainty of 190 kg h⁻¹ passes this criterion, yet the confidence interval spans a factor of approximately four. The authors should discuss whether comparisons with EPA emission factors (which carry their own uncertainty) remain meaningful at this level of DI uncertainty, and consider showing the distribution of relative uncertainties across the 89 robust plumes. Additionally, the standard deviation across the growing-box series is used as the uncertainty, but this captures only variability in the box-integration approach; it does not include wind field uncertainty (HRRR vs. actual on-site winds) or retrieval precision uncertainty. A brief discussion of error budget contributions, even qualitative, is needed.

4. Comparisons with prior studies (Table 1)

Table 1 compares mean MethaneAIR-derived per-animal emission rates with values from the EPA inventory, IPCC Tier 2, Golston et al. (2020), and McCabe et al. (2023). Several aspects of this comparison are problematic and require clarification. First, the Golston et al. (2020) and McCabe et al. (2023) measurements were conducted under specific seasonal and meteorological conditions, whereas the MethaneAIR data span multiple seasons and years; aggregating across these without accounting for seasonal variability introduces a systematic confounding factor. Second, McCabe et al. (2023) is described in a footnote as focusing primarily on beef feedlots but including a small number of dairy operations, yet the blended mean (13 ± 2 g animal⁻¹ h⁻¹) is placed in the table in a row shared with both CAFO types. This is potentially misleading and should be restructured. Third, the IPCC Tier 2 values listed (∼6.62 for beef and ∼13.95 for dairy) differ substantially from the EPA Colorado values (6.67 and 22.25 respectively); the reasons for this discrepancy should be explained.

5. Source attribution methodology and its limitations

The plume source attribution (Section 2.5) is described as a manual, qualitative process based on proximity, wind-consistent alignment, morphology, and absence of nearby competing sources. While this approach is pragmatic, the manuscript provides no inter-analyst reliability estimate (e.g., from duplicate independent evaluation of a subset of scenes) nor any false-positive rate estimate from the qualitative assignment. This limits the reader's ability to judge the confidence of the 200 agricultural plume attributions. The authors should provide at least some quantification of this uncertainty or whether a subset of scenes was evaluated by two independent analysts.

Minor Comments

6. Lines 103–106: DI detection threshold (500 vs. 120 kg h⁻¹)

The introduction states that the DI method was effective over 500 kg h⁻¹ in Chulakadabba et al. (2023), while Guanter et al. (2025) achieved 120 kg h⁻¹. The present study claims to detect plumes below these thresholds using scene-based subsetting. The authors should more explicitly quantify the effective detection threshold achieved in the present agricultural application, and explain mechanistically why scene subsetting enables detection at lower flux rates.

7. Figure 6 and Section 3.1: Detection threshold curves

Figure 6 shows detection rates as a function of estimated CAFO emissions and wind speed. The "estimated CAFO emissions" on the x-axis of the left panel appear to be derived from the same CDPHE permitted capacity data that is used as the denominator in per-animal calculations. If correct, this means the detection curve is not independent of the capacity data uncertainty. The authors should clarify what exactly is shown on the x-axis and how it was calculated.

8. Lines 257–260: Averaging of "robust, disconnected, non-unique plumes"

The procedure for computing total scene emissions is described as averaging robust disconnected non-unique plumes and summing remaining unique plumes. This distinction between "disconnected non-unique" and "unique" plumes should be more carefully defined earlier in the methods section, with a schematic or example to clarify how these categories are assigned and averaged.

9. Section 4, Lines 371–378: Overstated causal interpretation

The conclusion that dairy CAFOs exceed EPA estimates "likely due to higher operational intensity or greater contributions from manure management systems" is speculative without supporting evidence from the data themselves. The alternative explanation, that permitted capacity substantially underestimates actual capacity at dairy facilities, should be given equal weight in the interpretation, especially given the concern raised in major comment 1 above.

Technical Corrections

• Line 141: "were were" should be "were"
• Line 280: "6.67" in the text does not match "6.57" cited earlier (line 264); one of these appears to be a typographical error and should be made consistent with the EPA source.
• Line 360/362: "89 (42%)" is reported in the conclusions but the abstract and Section 3 state 44%; please reconcile.
• Line 377: "are be needed" should be "are needed"
• Line 185: "a the scene" should be "the scene"
• Figure 3 caption: The caption states "~0.005 degrees surrounding CAFO" but the methods describe a ±0.02° subset (Section 2.3). These dimensions should be made consistent.
• Line 115: McCabe et al. reference: the units "13 +- g of CH4 animal⁻¹ h⁻¹" appear incomplete; please supply the full value.
• Zhang et al. (2026): This reference (line 567–569) is to an EGUsphere preprint; if it is not yet published in a peer-reviewed journal, please clarify its status and ensure it is accessible to reviewers.

The paper uses a targeted, scene-based approach for detecting and quantifying methane emissions specifically from agricultural sources (CAFOs) using MethaneAIR data. There are several novel areas, including use of scene subsetting around known source locations and wavelet denoising to enhance detection, and the airborne application to agricultural sources. I generally agree with the existing reviews, with just a few additions:

Main

- How is a detection of None determined?

- How are wind fields handled, since the methane is a column-averaged quantity?

- Echo Referee 3 Comment #2 about estimates possibly being overestimated due to the robust plume filtering, which could bias towards stronger than average plumes

- Echo Referee 1 Comment #3 about the uncertainty budget being incomplete since it appears to be strictly based on repeatability across different box sizes

Minor

- Figure 6: It is somewhat unclear that ‘estimated’ here refers to the emission factor based estimate, not the MethaneAIR one. L306 even mentions detected emission rates. I was initially confused why emissions were estimated but then classified as a Non detect

- Figure 6: Is Farm ID part of the CDPHE data? It is not specifically mentioned

- Table 1: Goltson -> Golston