The authors quantify methane emissions from 12 US oil and gas basins using methane column observations from 32 MethaneAIR flights in 2023. These 12 basins accounted for 70% of total onshore oil and gas production in the contiguous United States in 2023. The authors estimate both total and sector-specific (oil + gas) emissions for each basin. They use a novel two-step regional flux inversion approach that first quantifies large point sources and then diffuse area emissions via Bayesian inverse analysis with the Stochastic Time-Inverted Lagrangian Transport (STILT) model. Emission contributions from non-oil and gas sources are estimated using sectoral emission estimates from a collection of previous top-down and bottom-up studies. The authors compare their regional estimates of methane emissions and loss rates with 16 previous studies and find generally good agreement.

The manuscript is well-written and a good fit for ACP. I recommend that it be accepted for publication with revisions to address the following comments and questions:

• I think a more detailed description of the flux inversion methodology is needed, ideally in the main text. It is a novel approach and the most critical part of the analysis. Can more information be provided? For example, it would be helpful to know more about the modeling of the “boundary inflow”, the numerical solution to the inverse problem, the approach to calculating column sensitivities with STILT (presumably based on MethaneAIR retrieval averaging kernels), and how/why GFS and HRRR meteorology are combined to drive STILT.
• What happens to the flux inversion if the modeled wind direction for a point source is wrong? In principle a flux dipole could arise, but it is mentioned in the SI that the non-negativity constraint in the inversion helps prevent that. How well does that work, and do any other biases manifest?
• How are point sources detected prior to the diffuse flux inversion? Is the process automated, semi-automated, or manual?
• 305-310: There is some redundant content in this passage.
• 314: 0.17 kg CH4/GJ from MethaneAIR is very similar to 0.18 kg CH4/GJ from IEA. Is it expected to be much lower? Perhaps this passage can be clarified.
• 319: I believe Figure 6 is mislabeled here––the passage seems to refer to Figure 5.
• Figures 5 and 6:
  • In which cases are the MethaneAIR and previous estimates for the same domain? Is the Zhang et al. result for the Permian spatially resampled to the flight domain? Those authors reported a loss rate of 3.7%, much higher than the <2% shown in Figure 6, so I assume so. It would be helpful to mark on the plots whether or not the previous results reflect spatial resampling.
  • Why are the x labels in the figures not identical? There are fewer bars in some subplots of Figure 6 than Figure 5.
  • Why do the inter-study differences in methane emissions not more closely match the differences in methane loss rate? Two examples of this: the Peischl bars in the Barnett subplots show much higher emissions than MethaneAIR but very similar loss rate, and the MethaneAIR bars in the Permian subplots show better agreement with previous studies for emissions than loss rate.

The authors present results from many MethaneAIR flights performed in the United States, primarily to quantify oil&gas emissions across major basins. Overall this represents a tremendous body of work to execute the flights, process data, and analyze for fluxes and the authors should be commended for this effort. The manuscript succinctly summarizes the results and shows general consistency in emission rates derived from previous observation studies (mass-balance, and satellite remote sensing). The authors employ what appears to be a novel way to calculate methane fluxes from remote sensing observations, but the details are extremely light. Before I can accept for publication, considerable more detail needs to be included and justified. My comments are as follows:

1. Table 1. Is Basin Area the total area of the basin or total area flown? If total area, can you express in the same table how much of that area you flew with MethaneAIR? 

Section S1. STILT. 

2. How do you simulate columns with STILT? How many layers? Interpolate between layers? Use an averaging kernel? What is the averaging kernel?

Section S1. The calculation of the background is unclear. 

3. Can you restate in terms of an equation, figure, or additional clarifying language? - "The background concentrations are given by a model..." - what model? STILT? 

4. "The boundary inflow is modeled using the Jacobian and emission rates outside the domain of observed concentrations." Where do you get these emissions? An inventory? Proper background quantification is so vital to robust inversions, this section needs to be much clearer.

Section S1. Point Sources.

5. Is the divergence integral method to calculate point sources applied at the 0.01 binning or at the native resolution? If you are binning, then you are certainly subtracting out more than point sources, as you are aggregating all true emission sources within that ~1km domain. 

6. If you are not binning, how do assess that model transport error correctly subtracts the influence of point sources? Do you have a quality control approach that ensures this? If the transport is wrong, then you risk not subtracting the point source component in your concentration field, which I can imagine will throw the inversion haywire and produce flux artifacts.

7. How do you determine an origin of the point source via the divergence integral method? It seems quite critical that you get the origin correct if you are are forward simulating a concentration field with STILT. 

8. It's not obvious to the reader that subtracting a forward model simulated concentration field produces a preferred result, especially assuming some level of spatial aggregation (e.g., were you to run these inversions at 10-20km, scales that others perform satellite inversions, would this still be required?). 

Section S1. Inverse Problem

9. What is the formulation of the inversion problem? An SI is a good place to put these equations to paper. The way it currently reads, the mass-balance constraint would just be the inverse of the Jacobian - i.e., s = (H^-1)y and the non-negativity constraint would be some sort of gradient descent (or something else?) algorithm that stops at zero. Not obvious from what's written. Are there other parameters that keep the solution from an overfit? They say there is no need for a prior, so not Bayesian I guess? 

10. The authors do not provide any sort of metric of goodness of fit (e.g., H * s_hat plotted against y) or information content from the retrieval. It's fairly common practice to show how well your model around the optimal emission state compares to observations. It's also common to show information content metrics (e.g., degrees of freedom for signal, model-resolution matrices, etc) for inversions, but given there's not an explicit inverse formulation in paper, it's hard to know if that would be feasible. 

11. Table S5. Please provide citations for literature based estimates, perhaps as an additional column in the table or a footnote.