Methane emissions inversions from MethaneSAT data produce 0.04°×0.04° (i.e., 4km×4km) resolution methane emission maps of total methane emissions at spatial scales of roughly 220km×440km from a single overpass. The satellite is equipped with a pair of Littrow passive imaging spectrometers that measure the column-averaged dry-air mole fraction of methane (i.e., XCH4) at a resolution of ∼110m×400m at nadir with a precision of 2.5–5.5 ppb at 2km×2km. Methane emissions are estimated from MethaneSAT column averaged mole fractions of methane (XCH4) using the Column Observations to Regional Emissions (CORE) inversion framework, which generates the MethaneSAT Level-4 emissions product (The MethaneSAT Science and Engineering Team, 2026). CORE relates observed methane columns to surface fluxes through a linear forward model z=Jintsint+Jextsext+(zprior+Ab) where z is the vector of observed XCH4 values, J is the Jacobian matrix describing the sensitivity of each observation to surface emissions, s represents methane emission rates within the interior and exterior domains, zprior is the prior methane column used in the Level-2 retrieval, A is the averaging kernel, and b is a background offset parameter.

The source–receptor relationship represented by J is computed using the Stochastic Time-Inverted Lagrangian Transport (STILT) model (Fasoli et al., 2018; Lin et al., 2003) driven by meteorological fields from the Global Forecast System (National Oceanic and Atmospheric Administration, 2026). STILT simulates backward particle trajectories from observation locations to quantify the sensitivity of each observation to upwind methane emissions. STILT footprints extend up to 28 h back in time from the satellite observation, which is the ventilation time scale for the observed region size in typical wind conditions, while assuming constant emissions during this time.

MethaneSAT Level-3 observations are aggregated to 2km×2km pixels prior to inversion to reduce measurement noise and computational cost. Emissions are estimated on a 4km×4km grid within an interior domain defined by contiguous regions of valid observations, while potential emission sources extending up to 300 km beyond the observed domain are included to represent inflow contributions through atmospheric transport. The background methane column is represented as the retrieval prior plus an additive offset parameter scaled by the averaging kernel derived from the Level-2 methane retrieval (Chan Miller et al., 2024). Exterior emission sources are clustered according to the similarity of their transport footprints to reduce the dimensionality of the inverse problem.

Model parameters are estimated using Bayesian inference with state vector θ=(sint,sext,b). Posterior samples are generated using the Stan probabilistic programming framework (Carpenter et al., 2017) with the No-U-Turn Sampler (Hoffman and Gelman, 2011), an adaptive Hamiltonian Monte Carlo algorithm (Neal, 2011). Observation uncertainty is represented by a constant standard deviation of 11 ppb, and emission rates are assigned lognormal prior distributions. Posterior mean emission rates provide the estimated flux for each grid cell, and uncertainty on the total dispersed area emissions is the 95 % confidence interval from the posterior distribution (n=4000), with an additional 20 % uncertainty added to account for assumed uncertainty in the static parameters in the input GFS weather data used for the inversions.

Individual MethaneSAT emissions estimates (i.e., a MethaneSAT scene or emissions map) represent methane emission estimates up to 28 h back in time and vary in their spatial dimensions depending on the viewing geometry of the satellite. Nadir viewing observations produce up to ∼220km wide scenes while off-nadir observations produce up to ∼440km wide scenes. The MethaneSAT platform had an agile observing mechanism with off-nadir viewing at up to 40° on one side of its observing track. We aggregate multiple MethaneSAT scenes together over the same regions to produce a spatially explicit estimate of methane emissions with increased temporal and spatial extents. To do this, we reproject all MethaneSAT emissions estimates onto a common global 0.04°×0.04° grid (Fig. S1 in the Supplement) using an equal-area weighting approach that preserves total methane mass among the individual scenes. Then, we average the methane emission rates over each cell for each overlapping scene and combine to produce an aggregated methane emission heatmap. We apply the same approach using the median during aggregation as an additional test of sensitivity and find broad consistency between the two approaches (Fig. S10). We also test the sensitivity of MethaneSAT aggregated emissions estimates to the removal of single scenes and find that estimates are robust among the six regions (Sect. S2, Fig. S12). We additionally consider any impacts from seasonal variability in our aggregated emissions estimates using 2022 monthly inversions data from TROPOMI (Pendergrass et al., 2025) to better compare to annual emissions estimates from top-down satellite inversions or bottom-up inventories (Sect. S2).

We perform intercomparisons of total methane emissions estimates to other spatially explicit methane emissions data from both bottom-up and top-down methodologies. In all intercomparisons, we match the spatial domains to ensure that the same regions are being compared. In addition to total methane emissions, we also produce comparisons of available literature-based emissions from different sectors (i.e., oil and gas and non-oil and gas emissions). From the bottom-up inventories, we specifically compare MethaneSAT to the EPA-GHGI (Maasakkers et al., 2023) for observations in the US, and to EDGAR – version “EDGAR_2025_GHG” (Crippa et al., 2024) and CAMS v6.2 (Granier et al., 2019) for regions outside of the US. The EPA-GHGI provides annual estimates of methane emissions for 2020, while EDGAR and CAMS v6.2 both report annual emissions for 2024. The EPA-GHGI and EDGAR provide emissions estimates for member countries under the UNFCCC. Other bottom-up inventories that are used as prior information for satellite-based inversions include the Global Fuel Exploitation Inventory (GFEI v2) (Scarpelli et al., 2022) and CAMS v6.2 (Granier et al., 2019). Our comparisons to top-down satellite-based observations vary by region and are presented later in the Results section.

We attribute MethaneSAT methane emissions to methane sectors by leveraging a combination of bottom-up inventories and Carbon Mapper detected point sources in a composite prior inventory using a simple proportional allocation, where ei are MethaneSAT emissions from sector i, E are total methane emissions estimates from MethaneSAT, pi,j are emissions estimates from prior inventory j and sector i, and Pj are total methane emissions from prior inventory j. ei=E∑jpi,j∑jPj To account for structural omissions in any single inventory, we construct a composite dataset as the sum of multiple independent bottom-up methane emission inventories (Fig. S5). This composite data approach has precedents in methane inversion frameworks (Cusworth et al., 2021a; Lu et al., 2023; Shen et al., 2023) where multisource priors or spatially-explicit inventories are used to improve completeness and robustness. For bottom-up inventories – we incorporate the gridded EPA-GHGI (Maasakkers et al., 2023), EDGAR (Crippa et al., 2024), EI-ME (Omara et al., 2024), CAMS v6.2 (Granier et al., 2019), and GFEI v2 (Scarpelli et al., 2022) as inputs, which are all mapped at their native spatial resolution of 0.1°×0.1°. For any region and sector, we combine emissions estimates from two bottom-up inventories with Carbon Mapper distinct point sources to produce the composite dataset. For regions within the US, we use the EPA-GHGI and EDGAR as inputs for non-oil and gas sources, and the EI-ME and EDGAR for oil and gas sources. For regions outside of the US, we use CAMS v6.2 and EDGAR as inputs for non-oil and gas sources, and the GFEI v2 and EDGAR as inputs for oil and gas sources. Other combinations of these bottom-up inventories, and their impacts on the resulting sectoral breakdown of emissions, are provided in Figs. S7 and S8 with further explanation in Sect. S1.1 in the Supplement. We include persistence-weighted distinct point source measurements from Carbon Mapper (Carbon Mapper, 2025) if they are sources that have been observed at least three times. To better account for regions where granular oil and gas infrastructure data is limited (i.e., regions outside of the US), we spatially aggregate oil and gas methane emissions estimates from the composite prior by a factor of four (i.e., from the native 0.1°×0.1° resolution to 0.4°×0.4°) while conserving the mass of emissions. A sensitivity test of this approach shows improved agreement in the sectoral proportions of emissions for non-US regions compared to a compilation of estimates from literature, while showing little to no improvement for regions in the US (Sect. S1.2).

We allocate total methane emissions estimates from MethaneSAT to broad sectors outlined as oil and gas (i.e., upstream and midstream sectors), waste (i.e., solid waste disposal landfills), agriculture (i.e., manure management, enteric fermentation, and agricultural soils such as rice cultivation), coal, and other non-oil and gas sources (i.e., post-meter emissions, wastewater treatment, chemical processing, etc). For cells where we find over 100-times the total methane emissions relative to the composite prior, we assign emissions as having an “unknown” origin and incorporate their relative percentage contributions into the uncertainty calculations related to sectoral disaggregation. Methane emissions from wetlands and termites are not included in the sectoral disaggregation since their combined methane emissions are less than 1 % of total methane emissions from the observed regions based on data from WetCHARTs (v1.3.1) (Bloom et al., 2017).

We acknowledge that a wide range of measures exist under the umbrella of methane intensity (Johnson et al., 2026; Seymour et al., 2025), so we assess oil and gas methane intensity using two distinct and complementary metrics:

Marketed gas-production-normalized methane intensity, defined as the ratio of oil and gas methane emissions to marketed methane production (i.e., loss rates).

Uncertainty in the MethaneSAT emissions product is dominated by: (1) uncertainty in the meteorological product used to generate the STILT Jacobian that links emissions with concentrations, (2) correlated uncertainty in the observations (e.g., striping), (3) uncertainty in the background concentration, (4) uncertainty in the allocation of signal between emissions in the reported domain and the boundary inflow, and (5) uncertainty in the emission map as expressed by the variability in samples in a Markov Chain Monte Carlo (MCMC) simulation. Uncertainty in aggregated MethaneSAT emissions estimates is propagated using a Monte Carlo approach. Each emissions cell is represented by 4000 samples drawn from its MCMC posterior distribution, reflecting mean-level uncertainty in the emissions estimate at that location. Where multiple emissions maps overlap a given cell, 4000 combined cell-level estimates are generated by repeatedly drawing one value per map and averaging across maps. This Monte Carlo resampling procedure propagates uncertainty through the arithmetic mean without requiring assumptions about the functional form of the resulting distribution. This procedure is applied independently to all subregions defined by unique combinations of overlapping emissions maps (Fig. S2). Uncertainty on the total dispersed area emissions is the 95 % confidence interval on the total across all samples (n=4000), with an additional 20 % uncertainty added to account for assumed uncertainty in the static parameters in the input GFS weather data used for the inversions.

To calculate uncertainties related to the disaggregation of methane emissions by sector, we bootstrap with resampling (n=4000) the input data used to create the prior emissions estimates in our stacked prior inventory, which are in turn used to re-calculate the disaggregation of methane emissions. For the bottom-up inventories, regardless of sector, we assume a normal distribution with a standard deviation of 50 % for the cell-level emissions estimates. For the Carbon Mapper point sources, we use the provided source-level standard deviations assuming a normal distribution to resample the emission rates (n=4000). Sectoral ratios of methane emissions are then calculated 4000 times using these resampled input data to provide upper and lower bounds on the uncertainty for the associated sectoral methane emissions, which we then use to obtain the 97.5th and 2.5th percentiles as the sectoral disaggregation uncertainty including the added relative contributions of unknown methane emissions to the total. The additional uncertainty relating to the attribution of unknown methane emissions from MethaneSAT contributes only <1% of uncertainty to the total for the regions we analyze in this work.

We present MethaneSAT observations from six distinct regions intersecting six countries, seven major oil and gas producing basins, and 207 districts/counties (i.e., Level 2 data from the Global Administrative Areas database – GADM). All regions are named according to the primary oil and gas basin encompassed by MethaneSAT observations (Table S1 in the Supplement), even if the full basin is not contained within the full observation domain.

Regions A, B, and C, are all located in North America, mostly in the United States (US). Region A (i.e., “Permian” observation domain) covers the oil-dominant Permian basin (i.e., >50% of combined energy production is from oil) (Fig. S9), one of the highest producing basins in the world with a long legacy of resource development and a sustained production surge beginning in the 2010's, leading to significant growth in associated infrastructure development (Scanlon et al., 2017). Region B (i.e., the “San Joaquin” observation domain) targets the state of California (US) encompassing regions with elevated methane emissions associated with a mixture of oil and gas activity, landfills, and livestock-related agricultural activity (Duren et al., 2019; Miller et al., 2015; Vechi et al., 2023). The San Joaquin basin is a mature oil-dominant basin (Fig. S9) with production dating back to the late 1800's, with many older wells presently active. Region C (i.e., the “Eagle Ford” observation domain) principally targets the Eagle Ford oil and gas basin in the US, but also extends into Mexico with some coverage of the Sabinas basin where coal production first began in the country (Dávila-Pulido et al., 2023). The Eagle Ford oil and gas basin is one of the youngest hydrocarbon basins we analyze in this work, with the first wells drilled in the late 2000's and an overall balance of oil versus gas production (Fig. S9).

Regions D, E, and F are all located in Asia and the Middle East. Region D (i.e., “Amu Darya – UZB”) covers most of the province of Qarshi (Uzbekistan) with some overlap into Turkmenistan. The Amu Darya – UZB region overlaps the Amu Darya oil and gas basin and encompasses multiple oil and gas fields along the border of Uzbekistan and Turkmenistan (Yu et al., 2015). Region E (i.e., “Amu Darya – TKM”) covers a separate portion of the Amu Darya basin that contains several major gas fields (Yu et al., 2015) and the city of Mary, the fourth largest city in Turkmenistan. The Amu Darya basin itself is predominantly gas-producing (i.e., >90% of combined production is gas) (Fig. S9). Region F (i.e., “Zagros Foldbelt”) targets the Zagros Foldbelt oil and gas basin in Iran, with partial coverage over the Widyan oil and gas basin in Iraq (Fig. S2). The Zagros Foldbelt basin is a large oil-dominant basin (Fig. S9) with a long history of oil and gas production dating back to the early/mid 1900's (Alipour, 2024).

Total methane emissions for the regional six observations domains are: 454 th-1 (95 % c.i.: 351–563) in the Permian, 251 th-1 (95 % c.i.: 189–321) in the Zagros Foldbelt, 192 th-1 (95 % c.i.: 146–242) in Amu Darya – UZB, 188 th-1 (95 % c.i.: 141–239) in Amu Darya – TKM, 127 th-1 (95 % c.i.: 95–162) in the San Joaquin, and 114 th-1 (95 % c.i.: 83–149) in the Eagle Ford (Fig. 1).

We attribute methane emission estimates from MethaneSAT to specific methane sectors and find varied sectoral emissions among the different observation domains, reflecting a diversity of methane emitting sources among the different regions. The Permian contains the highest percentage of oil and gas methane emissions at 90 % (95 % c.i.: 64 %–100 %), followed by the Zagros Foldbelt at 81 % (95 % c.i.: 58 %–100 %), the Eagle Ford at 70 % (95 % c.i.: 41 %–99 %), Amu Darya – UZB at 52 % (95 % c.i.: 37 %–70 %), Amu Darya – TKM at 52 % (95 % c.i.: 38 %–66 %), and the San Joaquin at 24 % (95 % c.i.: 18 %–31 %) (Fig. 2). Among non-oil and gas sources, the dominant sector was consistently agricultural emissions associated with livestock like concentrated animal feeding operations (i.e., CAFO's) and manure management (Fig. 2). After the agricultural sector, non-oil and gas emissions from the waste and other (i.e., wastewater treatment, post-meter, stationary combustion, etc) sources were the most prominent emission sectors in Amu Darya – UZB, Amu Darya – TKM, Zagros Foldbelt, and San Joaquin (Table S2). For the Eagle Ford region, the waste and coal sectors were the highest methane-emitting sectors from non-oil and gas sources after the agricultural sector. Detailed sectoral emissions estimates are shown in Table S2.

We calculate methane intensities based on marketed gas (i.e., loss rates) and total oil and gas (i.e., energy intensity) production for 2024 (Wood Mackenzie, 2025) from all aggregated domains and for notable administrative/sub-basin boundaries among the observation domains (Fig. 3). Loss rates among the six regions consistently exceed the 0.2 % goal set within the Oil and Gas Climate Initiative by factor of ten (OGCI, 2025). The Eagle Ford region has the lowest loss rates at 2.4 % (95 % c.i.: 1.4 %–3.4 %), followed by the Permian at 2.6 % (95 % c.i.: 1.9 %–3.5 %), Amu Darya – TKM region at 2.9 % (95 % c.i.: 2.1 %–3.7 %), Amu Darya – UZB region at 3.7 % (95 % c.i.: 2.7 %–5.0 %), the San Joaquin region at 12.1 % (95 % c.i.: 8.9 %–16.0 %), and the loss rates in the Zagros Foldbelt at 18.6 % (95 % c.i.: 13.4 %–23.8 %). For energy intensities, which accounts for combined marketed oil and gas production, we estimate the lowest energy intensities for the Zagros Foldbelt at 0.16 (95 % c.i.: 0.12–0.20) kgCH4GJ-1 and the Permian at 0.16 (95 % c.i.: 0.12–0.21) kgCH4GJ-1, followed by the Eagle Ford at 0.21 (95 % c.i.: 0.12–0.30) kgCH4GJ-1, the Amu Darya – TKM region at 0.40 (95 % c.i.: 0.29–0.51) kgCH4GJ-1, the San Joaquin region at 0.42 (95 % c.i.: 0.31–0.55) kgCH4GJ-1, and the highest energy intensities in the Amu Darya – UZB region at 0.51 (95 % c.i.: 0.27–0.69) kgCH4GJ-1. We observe similar intensity metrics, both for loss rates and energy intensities, within the specific spatial domains of oil and gas basin boundaries compared to the full observation domains (Fig. 2 and Table S3).

We compare methane emissions estimates and methane intensities across sub-basins and administrative boundaries (e.g., countries/states) (Fig. 2). Within the Delaware and Midland subbasins of the Permian (Fig. 2), we estimate oil and gas methane emissions of 178 th-1 (95 % c.i.: 132–224) and 112 th-1 (95 % c.i.: 81–143) respectively, with comparable marketed loss rates of 2.0 %. The Permian basin transects both the New Mexico and Texas state boundaries, where we estimate oil and gas methane emissions of 67 th-1 (95 % c.i.: 43–90) and 338 th-1 (95 % c.i.: 249–427) respectively. We find over twice the loss rates and energy intensities values in Texas at 3.1 % and 0.19 kgCH4GJ-1 compared to New Mexico at 1.3 % and 0.08 kgCH4GJ-1 (Fig. 2 and Table S3). The same state-level comparison restricted to the Delaware subbasin boundary shows a similar contrast in methane intensities with the Texas portion of the Delaware subbasin having a loss rate of 2.8 % compared the New Mexico portion of the Delaware subbasin at 1.0 %. Similar trends have also been observed in recent TROPOMI-based estimates by Varon et al. (2026), with New Mexico showing decreasing loss rates from 4.5 % in 2019 % to 2.1 % in 2023 plausibly associated with state-wide policies requiring operators to reduce methane intensities below 2 % by 2026 (N.M. Code R. § 19.15.27.9 Statewide Natural Gas Capture Requirements, 2025).

In the Eagle Ford, we note that nearly all oil and gas emissions occur within the US compared to Mexico, with methane emissions in Mexico largely originating from agricultural sources and coal (Fig. 2 and Table S6). Nearly all oil and gas infrastructure is located within the Eagle Ford oil and gas basin boundary, with sparse infrastructure in Mexico compared to the US (Omara et al., 2023). MethaneSAT observations in Mexico largely transect the Burgos basin, a major natural gas-producing basin. Although geologically similar to the Eagle Ford basin, daily gas production in the Eagle Ford exceeds 7000 MMcfd-1 compared to 30 MMcfd-1 from the Burgos, highlighting stark differences in the degree of development between the basins which we can clearly observe in the oil and gas methane emission estimates from MethaneSAT. We find that the methane intensities in the Zagros Foldbelt oil and gas basin in Iran (i.e., 25.5 % and 0.24 kgCH4GJ-1) are over five-times higher than the bordering Widyan oil and gas basin in Iraq (i.e., 4.2 % and 0.03 kgCH4GJ-1), noting that MethaneSAT observations only cover 17 % of combined oil and gas production in the Widyan basin compared to 58 % from the Zagros Foldbelt basin (Fig. S9). Within the Amu Darya basin boundary, we find higher methane intensities from the Amu Darya – TKM region at 4.1 % and 0.57 kgCH4GJ-1 compared to the Amu Darya – UZB at 3.6 % and 0.49 kgCH4GJ-1 (Table S3). Collectively, MethaneSAT observations from the Amu Darya basin have an associated loss rate intensity of 3.8 % and an energy intensity of 0.53 kgCH4GJ-1.

Our estimates of methane emissions and sectoral breakdowns from MethaneSAT match closely with other independent estimates for the same spatial domains from other satellite observations (Fig. 3). The Permian basin has been extensively surveyed using a wide range of methods from ground-based surveys (Robertson et al., 2020; Yu et al., 2022), tower-based observations (PermianMAP, 2025; Barkley et al., 2023), aerial-based surveys (Chen et al., 2022; Cusworth et al., 2021b; Hmiel et al., 2023; Sherwin et al., 2024), and satellite-based observations (Cusworth et al., 2022; Irakulis-Loitxate et al., 2021; Lu et al., 2022; Nesser et al., 2024; Shen et al., 2022; Varon et al., 2026; Worden et al., 2022). We find that our estimate of oil and gas emissions in the Permian of 408 th-1 (95 % c.i.: 303–516) for 2024 are similar to recent TROPOMI inversions by Varon et al. (2026) and East et al. (2025), and generally higher than older satellite-based estimates from 2020 and 2019 (Fig. 3). Our estimated marketed gas-production normalized methane intensity within the Permian basin domain of 2.5 % (95 % c.i.: 1.9 %–3.1 %) for the Permian closely aligns with recent estimates from MethaneAIR for 2023 at 2.4 % (95 % c.i.: 1.5 %–3.2 %) (MacKay et al., 2026) (Table S3), noting that methane intensities from MethaneAIR are calculated using gross gas production instead of marketed gas production. Our estimate of non-oil and gas emissions in the Permian region of 41 th-1 (95 % c.i.: 26–75) closely matches recent TROPOMI based inversions from 2023 (East et al., 2025; Varon et al., 2026), but higher than older estimates from GOSAT (Lu et al., 2023; Worden et al., 2022).

In the Eagle Ford region, we find that our total emissions estimate of 114 th-1 (95 % c.i.: 83.6–150) is higher than most other independent satellite-based observations (Fig. 3) with differences largely attributable to higher oil and gas emission estimates from MethaneSAT. Our estimated loss rates within the Eagle Ford of 1.9 % (95 % c.i.: 1.3 %–2.8 %) closely matches recent MethaneAIR measurements from 2023 of 2.0 % (95 % c.i.: 1.6 %–2.7 %) (MacKay et al., 2026), noting our use of marketed versus gross gas production. We find no strong seasonal biases in our methane emissions estimates for the Eagle Ford (Sect. S2, Table S6), implying that the differences between top-down estimates are reflective of the observed methane emissions.

In the San Joaquin region, we estimate 30 th-1 (95 % c.i.: 20–41) from oil and gas sources which compares well with multiple independent satellite-based estimates (Fig. 3). We find elevated loss rates within the San Joaquin oil and gas basin boundary at 15.5 % (95 % c.i.: 11.4 %–19.6 %), which is also observed in Omara et al. (2024) with a loss rate of 15.3 % for the entire San Joaquin oil and gas basin in 2021. The San Joaquin basin is characterized by a large proportion of marginally-producing well sites (Omara et al., 2018), which are typically associated with increased methane loss rates (Omara et al., 2022). We find higher emissions from non-oil and gas sources (i.e., 97 th-1) within the San Joaquin region compared to older satellite-based estimates (Lu et al., 2023; Worden et al., 2022), additionally noting that the aggregated MethaneSAT estimates may underestimate emissions by ∼10% based on the seasonal timing of the scene collections (Sect. S2, Table S6). We further note that our measurements were performed during daylight hours, which would also influence comparisons with annual average inventory estimates for the San Joaquin (Sect. S2).

In the Amu Darya – UZB region, our estimates of oil and gas emissions at 100 th-1 (95 % c.i.: 57–137) support the higher range of 39–110 th-1 (Fig. 3) found in independent satellite-based estimates. Oil and gas production within Uzbekistan has stabilized/declined in past decades (IEA, 2025b), supporting similar oil and gas methane emission estimates from past satellite-based inversions to estimates from MethaneSAT (Fig. 3). Our estimates of non-oil and gas emissions from the Amu Darya region of 92 (95 % c.i.: 55–134) th-1 exceed those from independent estimates which range from 13–39 th-1 (Fig. 3). Our total methane emission estimates from the Amu Darya – UZB region are also higher than other independent estimates, albeit with statistical overlap for estimates from GOSAT estimates in North Africa (Western et al., 2021) and TROPOMI estimates in the Middle East and North Africa (Chen et al., 2023). Annual trends for methane emissions in Uzbekistan, as reported to the UNFCCC from 1990–2012 (UNFCCC, 2025), indicate increasing emissions from non-oil and gas sources like enteric fermentation, landfills, and wastewater treatment versus declining/stable emissions from the energy sector. Estimates from MethaneSAT for the Amu Darya – UZB region may indicate a continuation of these trends.

In the Amu Darya – TKM region, we find close agreement to recent TROPOMI inversions from 2023 for both oil and gas and non-oil and gas emissions (East et al., 2025). Only one other satellite-based estimate contains full sectoral emissions for the region (Worden et al., 2022), which finds negligible non-oil and gas emissions (Fig. 3). Non-oil and gas emissions from the Amu Darya – TKM region are predominantly from agricultural and waste sectors respectively at 60 th-1 (95 % c.i.: 38–85) and 24 th-1 (95 % c.i.: 15–32) located in the North of the observation domain over the city of Mary (Fig. 1). Our estimates of loss rates in the Amu Darya – TKM region of 2.6 % (95 % c.i.: 1.7 %–3.6 %) are lower than the 4.9 % estimated from satellite-based observations for 2019 (Chen et al., 2023), although the MethaneSAT observations exclude the South Caspian area, a region that has been repeatedly observed with large point source emissions associated with oil and gas infrastructure (Irakulis-Loitxate et al., 2021; Varon et al., 2021).

Our total methane emission estimates from the Zagros Foldbelt region overlap with multiple satellite-based observations (Fig. 3), although we estimate higher emissions attributable to non-oil and gas sources compared to other satellite-based estimates that provide comprehensive sectoral disaggregation of methane emissions for the region (East et al., 2025; Worden et al., 2022). Our estimates of oil and gas emissions are within the range of other satellite-based estimates for the region (Fig. 3). We find high loss rates in the Zagros Foldbelt oil and gas basin at 25.5 % (95 % c.i.: 18.5 %–32.6 %), which is over ten-times higher than country level estimates of 0.8 % for Iran from satellite-based observations from 2019 (Chen et al., 2023). A comparison of methane emissions estimates within the MethaneSAT observation domain from the same study (Chen et al., 2023) are less than half the emissions estimated by MethaneSAT, which could contribute to the observed differences in methane intensities, in addition to other factors (i.e., variations in the spatial representation of the loss rate estimates, production characteristics, sectoral disaggregation methods, and study year).

In the US, we find that MethaneSAT observations from 2024–2025 are consistently higher than 2020 estimates from the gridded EPA-GHGI by a factor of 2–5. This finding echoes recent results from comprehensive aerial sampling campaign from MethaneAIR (MacKay et al., 2026), and a broader trend of top-down observations exceeding estimates from bottom-up inventories (Saunois et al., 2025). We find that both oil and gas, and non-oil and gas emissions estimates from MethaneSAT exceed those from the gridded EPA-GHGI, highlighting broad discrepancies among multiple methane-emitting sectors. Oil and gas emissions estimates from MethaneSAT are four-times higher compared to the gridded EPA-GHGI in the Permian and Eagle Ford regions and two-times higher in the San Joaquin. Emissions from the waste sector are consistent between MethaneSAT and the gridded EPA-GHGI, with most differences from non-oil and gas sources occurring from the agricultural sector with MethaneSAT finding higher estimates by a factor of four in the Eagle Ford and Permian, and a factor of two in the San Joaquin. For all three regions in the US, oil and gas emissions were consistently higher than the EPA-GHGI by a greater degree when compared to non-oil and gas emissions.

For regions outside of the US, we compare MethaneSAT observations to bottom-up emission inventories from EDGAR (Crippa et al., 2024) and CAMS v6.2 (Granier et al., 2019). For the Zagros Foldbelt, we find closer agreement to CAMS v6.2 compared to EDGAR. In Amu Darya – UZB, we find comparable estimates of oil and gas emissions from EDGAR and CAMS v6.2, but our estimates of non-oil and gas emissions are twice as high as the bottom-up estimates, largely due to increased emissions related to agriculture. We see the largest discrepancies between MethaneSAT and bottom-up inventories in the Amu Darya – TKM region, with MethaneSAT estimates of methane emissions more than five-times higher than the 33 and 22 th-1 estimated within CAMS v6.2 and EDGAR respectively (Fig. 4). Persistence-adjusted point source detections from Carbon Mapper alone amount to 20 th-1 in the Amu Darya – TKM region (Fig. S4), implying that bottom-up estimates are likely underestimating emissions in the region.

We quantify methane emissions from MethaneSAT for jurisdictions (i.e., second-level administrative divisions – county/districts) from all six regions analyzed in this work. We investigate differences between bottom-up inventory estimates within jurisdictional bounds using the high-resolution data provided from MethaneSAT observations.

In the Permian region, the five highest emitting counties (i.e., Reeves, Eddy, Culberson, Lea, and Pecos) collectively contribute 41 % of total methane emissions from the region (Fig. 5). We find that total methane emissions estimates are consistently higher than the gridded EPA-GHGI across multiple counties/districts, with the difference driven by oil and gas emissions (Fig. 5). In Reeves County (Texas), where we observe the highest total methane emissions in the region, estimates are roughly 10-times higher than the gridded EPA-GHGI national inventory. In Eddy and Lea counties, both located in the state of New Mexico, the bottom-up inventory differences are less pronounced at a factor of two. These trends also reflect our findings of methane intensities between the states of Texas and New Mexico (Fig. 2), where intensities relative to gas and oil and gas production in Texas over twice those in New Mexico. Discrepancies between the EPA-GHGI and MethaneSAT in non-oil and gas emissions are less pronounced in the Permian, which also reflects the dominance of oil and gas emissions.

In the Eagle Ford region, the five highest emitting counties (i.e., Webb, Dimmit, Lasalle, Maverick, and McMullen) cumulatively account for 63 % of total methane emissions within the Eagle Ford region (Fig. 5). MethaneSAT data shows consistently higher emissions compared to the EPA-GHGI in the US, and to EDGAR in Mexico (Fig. 5). The three highest emitting counties, Webb, Dimmit, and LaSalle, all have emissions estimates that are 3–4 times higher than the EPA-GHGI. By contrast, MethaneSAT emissions estimates in Maverick County are nearly 20-times higher than the EPA-GHGI, driven by differences in both oil and gas and non-oil and gas methane emissions. All three of the counties located within Mexico contain negligible emissions estimates from EDGAR. The highest emitting county we analyzed within Mexico is Sabinas County located within the Sabinas basin, Mexico's largest coal-producing region (Dávila-Pulido et al., 2023). Several distinct point sources detected by Carbon Mapper and IMEO-MARS (i.e., satellite: EMIT – NASA) attributable to coal emissions are also contained within Sabinas County (Fig. S3), with point source emission rates detected by EMIT-NASA ranging from 1.4–4.4 th-1, and Carbon Mapper reporting a persistence-adjusted methane emission rate of 1.8 th-1 (95 % c.i.: 1.5–2.1), similar to total methane emissions attributable to coal sources from MethaneSAT for this region of 1.6 th-1 (95 % c.i.: 1.1–2.2).

In the San Joaquin region, the five highest emitting counties measured by MethaneSAT (i.e., Tulare, Kings, Kern, Fresno, and Madera) account for 84 % of total methane emissions within the San Joaquin region. All three counties show increased emissions relative to the EPA-GHGI by a factor of 2–3. The MethaneSAT observation domain in the San Joaquin Valley encompasses a mixture of oil and gas and non-oil and gas methane emissions sources, with predominantly non-oil and gas methane emissions focused in Kings and Tulare counties, and mixture of oil and gas, agriculture, and waste emissions in Kern County (Fig. 5). The degree of difference between the EPA-GHGI and MethaneSAT estimates in San Joaquin region is lower than the Eagle Ford or Permian, potentially due to the relative lack of oil and gas methane emissions which has been highlighted as a major sector responsible for discrepancies between top-down and bottom-up estimates in the US (Alvarez et al., 2018). The diversity in the sectoral contributions of methane emissions in the San Joaquin region is also seen in the mapping of distinct point sources from IMEO-MARS and Carbon Mapper (Fig. S3), showing two clear regions of dense point source detections related to concentrated animal feeding operations (CAFO's) in Kings and Tulare counties and point sources related to oil and gas emissions in Kern County (Fig. 5). Multiple studies have highlighted the prominence of emissions from dairy sources in Kings and Tulare counties (Cui et al., 2017; Duren et al., 2019; Heerah et al., 2021; Miller et al., 2015). Despite rising milk production in the Tulare county region, the number of dairy operations dropped since the 1990's, reflecting structural changes to the dairy industry in California like the enlargement of herd sizes and consolidation of smaller farms into larger operations (Barrowman et al., 2025). Most oil and gas emissions estimated by MethaneSAT follow a semi-circular pattern enveloping the southeastern edge of the San Joaquin oil and gas basin, a pattern also observed in other spatially-explicit methane emission estimates like the EI-ME and GFEI v2. This portion of the San Joaquin oil and gas basin corresponds to a relatively dense area of oil and gas infrastructure including refineries and processing plants.

Methane emissions are evenly distributed among jurisdictions within the Amu Darya – UZB region, with the five highest emitting counties (i.e., G'uzor, Nurobod, Usmon Yusupov, Muborak, and Buzoro) collectively emitting 34 % of emissions for the entire region (Fig. 5). Comparisons of total methane emissions estimates from MethaneSAT to bottom-up estimates from EDGAR show variable discrepancies in emissions estimates (Fig. 5). Broadly, we find that MethaneSAT estimates higher methane emissions in the North in Nurobod and Buxoro but finds similar emissions to EDGAR in the South in G'uzor and Usmon Yusupov. The southern portion of the MethaneSAT observation domain contains most of the known oil and gas infrastructure in the region (Omara et al., 2023), including multiple distinct point source detections from IMEO-MARS and Carbon Mapper (Fig. S4). Increased methane emissions in the North are the primary explanation for why our overall estimates of methane emissions in the Amu Darya – UZB region are higher than multiple independent observations from both satellite and bottom-up inventories (Figs. 2 and 3).

Over half (i.e., 64 %) of total methane emissions in the Amu Darya – TKM region originate from the five highest emitting districts of Murgap, Sarahs, Baýramaly. Türkmengala, and Sakarçäge (Fig. 6). We consistently find elevated emissions estimates from MethaneSAT compared to bottom-up estimates from EDGAR (Fig. 6) to a greater degree than any of the regions we analyzed in this work. In Sarahs District, total emissions estimates from MethaneSAT at 32 th-1 are nearly 60-times higher than emissions estimates from EDGAR, due to elevated oil and gas methane emissions estimated by MethaneSAT. Across four of the five highest emitting districts/cities in the Amu Darya – TKM region, emissions estimate from MethaneSAT are a factor of 10–60 times higher than EDGAR (Fig. 6), with the exception being Türkmengala District. We also note broad discrepancies observed in the northern portion of the Amu Darya – TKM region in between Baýramaly and Sakarçäge and around the city of Mary (Fig. 6). In this region, we find a high density of methane hotspots detected by TROPOMI – Sentinel-5P (Schuit et al., 2023), but a lack of point source detections from other instruments (i.e., Carbon Mapper, EMIT, PRISMA) (Fig. S4) potentially caused by limited instrument targeting in this area or by higher emissions dispersed across wider areas that may be below the detection limit of high-emitting point source detection instruments.

Across districts in the Zagros Foldbelt region, we find consistent agreement, and even underestimation, of methane emissions from MethaneSAT compared to EDGAR (Fig. 6). In Iran, top-down inversion studies have reported methane emissions lower than EDGAR and closer to UNFCCC inventories (Maasakkers et al., 2019), with discrepancies likely arising in part from differences in the representation of oil and gas emissions and EDGAR's use of generalized emission factors (Crippa et al., 2024). Cumulatively, the top five highest emitting districts (Ahvaz, Ramhormoz, Behbahan, Kohgiluyeh, and Gachsaran) are responsible for 66 % of total methane emissions in the Zagros Foldbelt region. Except for Kohgiluyeh, MethaneSAT emissions estimates for all these jurisdictions are within a factor of 2 compared to EDGAR. Residual emissions in the region show neighboring positive and negative values, especially for oil and gas emissions following the NE-SW domain of the Zagros Foldbelt basin (Fig. 6).