the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Sep 2026
Quantifying total methane emissions in the Permian Basin using MFIM: the MethaneSAT FLEXPART Inverse Model
Abstract. MethaneSAT was a spaceborne imaging spectrometer that observed column-averaged dry-air mole fractions of methane (XCH4) in the atmosphere from March 2024 to June 2025. The data have fine spatial resolution (110 m x 400 m at nadir), high precision (~3 ppb at 2 km x 2 km aggregation) and wide swath (220 km at nadir). The goal of MethaneSAT is to provide comprehensive emissions characterization for oil and gas production regions, spanning the gap between area flux mappers and point source imagers. In this paper we introduce the MethaneSAT FLEXPART Inverse Model (MFIM), an inverse modeling framework for quantifying total emissions and their spatial distribution from MethaneSAT observations. We use a Lagrangian particle dispersion model, the FLEXible PARTicle dispersion model (FLEXPART v.11), in time reversed mode to determine the influence of surface emissions on atmospheric concentrations across a MethaneSAT scene (aggregated to 0.02° x 0.02° resolution). We then obtain the optimum emissions field using a Hamiltonian Monte Carlo (HMC) sampler to solve the inverse problem. We apply the inversion system to 9 scenes from the Permian Basin, yielding an annual emission rate of 3.49 ±1.43 Tg/yr (398 ±163 tons/hr). Emission estimates from MFIM resolve high emitting regions within the basin without the use of a spatially informed prior. MFIM leverages the strengths of the MethaneSAT observing system (104 independent observations) to produce spatially explicit, high-resolution (0.05° x 0.05°) emission estimates for application to monitoring and mitigation of methane emissions. Our results compare well to the MethaneSAT Level 4 product from the operational CORE (Column Observations to Regional Emissions) inversion algorithm, as well as other top-down studies in the region, lending confidence that regional methane flux estimates from MethaneSAT are robust against differences in modeled transport.
- Preprint
(2777 KB) - Metadata XML
-
Supplement
(29298 KB) - BibTeX
- EndNote
Status: open (until 29 Oct 2026)
- RC1: 'Comment on egusphere-2026-4693', Anonymous Referee #1, 14 Oct 2026 reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 326 | 132 | 43 | 501 | 65 | 47 | 41 |
- HTML: 326
- PDF: 132
- XML: 43
- Total: 501
- Supplement: 65
- BibTeX: 47
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The authors present MFIM (MethaneSAT FLEXPART Inverse Model), a Bayesian inversion of column-averaged methane (XCH4) from MethaneSAT. It uses FLEXPART backward trajectories, weighted with averaging kernels, following recent implementation by Thompson et al. in FLEXINVERT. It is applied to 9 scenes over the Permian Basin. The composite gives 398 ± 163 t/h (3.49 ± 1.43 Tg/yr). The authors compare it with the operational CORE (Column Observations to Regional Emissions) inversion, which uses STILT (Stochastic Time-Inverted Lagrangian Transport), with EI-ME (the spatially explicit Emissions Inventory of Methane Emissions), with a box model, and with published top-down studies. They conclude that MethaneSAT flux estimates are robust against transport differences.
Although well written and well structured, the manuscript as submitted requires significant improvements, mainly with regards to scientific significance and demonstration of claims.
The comparison of MFIM with CORE is presented as the cornerstone of the paper and MFIM is presented as an improvement from CORE but present arguments are not sufficient to support the claim.
Major revisions are needed, including native runs and experiments, further uncertainty tests, update of the error modelling, as well as further discussion of errors coming from transport, background, error statistics, etc.
General comments
1. Bibliography and references
The bibliography used in the paper is highly Americano-centric and does not fully acknowledge methane estimate studies, especially done with satellites, in e.g. Europe, Japan, India and China. Please broaden and rebalance the references, especially in the introduction.
2. Comparison with CORE
A large part of the manuscript is based on comparing with CORE. First, the reader needs to have all essential information about CORE to make this paper stand-alone without looking at other references. Second, the comparison itself seems to favor CORE, which has a more elaborate handling of errors and background. This should be further discussed, commented and assessed, in particular with specific native synthetic experiments.
3. Transport errors
As stated in the manuscript, transport errors are the dominant factor of uncertainty. This should be further assessed.
OSSEs with varying meteorological fields should be tried.
Correlation of errors should be accounted for a transport errors are correlated and are the dominant source of uncertainty.
Moreover, the difference of resolution between observations, transport and control vector smoothes out most of the observed signal (as stated in the manuscript).
This should be further discussed and tested, by e.g. aggregating Jacobian to observed aggregates at the same resolution.
4. Overall budget
The yearly budget is extrapolated from 9 scenes over the year. Errors are propagated somehow, but the yearly budget uncertainty needs to be further discussed.
But then, with such high uncertainty of the yearly budget, one could ask what is the added value of MFIM (or CORE) compared to the simple mass approach.
This needs to be further discussed, especially with regards to existing and future observing capability from other constellations (e.g., Tanager, GHGSat, TANGO, CO2M, Sentinel 5, etc.)
Specific comments:
- line 49: Add references about the average emission rate of different bassins, showing the variety of efficiency depending on the country and main operator.
- line 70: You base your article on a comparison with the CORE system. To make the manuscript stand-alone, add further description of CORE
- line 53: Make it clear from the beggining that you are using aggregates in the inversion system and not single native pixels
- line 205: All this method to handle the PBL height seems unecessary tricky. FLEXPART gives explicit altitude of individual altitudes, which makes it easy to filter out what particle is in the PBL. Also why interpolating based on a sigmoid when you have 3-hourly? How good does it match with the overall pattern? Do you use a minimal threshold at night for the PBL? please check with classical FLEXPART-based inversion frameworks (mainly FLEXINVERT) the consistency of your approach. Overall, your vertical gridding and PBL height interpolation adds some appoximation that are probably not critical compared to other sources of uncertainty, but not necessary... FLEXPART natively computes surface sensitivity that could be used directly.
- line 234: I do not understand this sentence
- line 234: You need to clarify the resolution of the processing chain. At what resolution do you aggregate FLEXPART footprints? You run backward trajectories for 2x2km boxes, but with 25km meteorological data; the gap of resolutions will likely produce artefacts, especially as the variability of 2x2km MethaneSAT observations is then aggregated at larger scale, loosing the fine scale information of the observations. Please make a test of the impact of the resolution. Intuitively, it seems that aggregating MethaneSAT observations directly at the control space resolution would lead to same results at much lower cost. This is actually what Figure 4 tells you. And what you comment line 410.
- line 250: The title is misleading. Replace by e.g., "background estimation"
- line 321: Using Normal distributions for the errors breaks the mathematical framework. Chi2 or log-normal distributions are much more realistic and relevant for these applications. Please check Ganesan and following related references for proper definition of the framework.
- line 360. You need to make a thorough estimates of the computational costs of your method, including Jacobian computations and Markov Chain walks. This is critical to comment the generalization of your methods to other bassins, even globally, and to other instruments, spanning more than a year, and in the future with possibly higher resolutions
- line 390: the header of section 6 is very long and should be included as a subsection itself
- Section 6.2: The comparison seems to point at more convincing results for the CORE algorithms, most likely due to the log-normal hypothesis on error distribution. Also, the transport resolution most likely plays a significant role on the quality of the results
- line 690: This claim seems over-stated. TROPOMI overpasses (and soon GOSAT-GW, Sentinel 5 and CO2M) would likely quickly converge to a reasonable estimate of the bassin budget.