For each MethaneAIR flight domain, we perform bootstrap resampling with replacement (n = 5000) of previous estimates selected from the studies listed in Table 2 to develop ranges (i.e., mean and 95 % confidence intervals) of oil and gas, non-oil and gas, and total methane emissions for the area. Briefly, we define oil and gas emissions as those originating from activities and infrastructure involved in the production, processing, transport, and distribution of oil and natural gas (e.g., well sites, pipelines, compressor stations, natural gas processing plants). Non-oil and gas emissions are any emissions from coal, agricultural (e.g., livestock, manure management), waste (e.g., landfills, wastewater treatment), and other industrial sectors. For domains with less than four unique estimates, we use the minimum and maximum estimates from previous studies as the range. In some cases, bottom-up data (e.g., Crippa et al., 2024; Maasakkers et al., 2023) are used to inform the ranges of non-oil and gas methane emissions due to a lack of measurement-based data (see Sect. S2 for additional discussion). We further use the ranges for non-oil and gas methane emissions to estimate the relative sector (i.e., oil and gas, non-oil and gas) contributions of total emissions quantified by MethaneAIR.

We estimate total oil and gas methane emissions (Eq. 1) as the difference between the total methane emissions quantified by MethaneAIR, MethaneAIRtotal, and an estimate of total non-oil and gas methane emissions in the MethaneAIR observation domain based on non-oil and gas methane emission estimates from previous literature, LEmisnon-ong. We also subtract any point source emissions quantified by MethaneAIR attributed to non-oil and gas sources, Pt_MAIRnon-ong), which were observed in five out of 12 basins. 1MethaneAIRong=MethaneAIRtotal-(LEmisnon-ong+PtMAIRnon-ong) For the above method, we acknowledge that subtracting a literature-based estimate of non-oil and gas methane emissions as well as non-oil and gas point source emissions quantified by MethaneAIR may introduce double counting as some of the MethaneAIR point source emissions may also be captured in literature-based estimates. However, since bottom-up data is included in many of the literature-based estimates for these regions, it is possible that the estimate could be low-biased, since recent research has shown that methane emissions from sectors such as waste are being under reported in bottom-up inventories (Cusworth et al., 2024; Moore et al., 2023; Nesser et al., 2024). If this is the case, the potential issue of double counting could be negligible, although difficult to confirm with current limited empirical data on non-oil and gas emissions. To help us understand whether this approach is reasonable, we also explore another method to estimate the relative sector contributions of methane emissions for regions of interest. Our alternative method utilizes spatially explicit methane emissions data for oil and gas sources from the measurement based inventory developed by Omara et al. (2024), updated using 2023 activity data, and non-oil and gas sources from the EPA 2020 GHGI (Maasakkers et al., 2023) to estimate grid-level ratios of oil and gas emissions, which are then applied to the quantified area emissions from MethaneAIR retrievals. This additional approach is further discussed in Sect. S2, with detailed comparisons of estimates derived by the two methods shown in Fig. S2.

MethaneAIR data has sufficient spatial resolution and precision to disaggregate emissions by sector in many basins, and sufficient swath to determine regional totals. Robust, observation-based disaggregation of methane emissions across sectors is an important part of an actionable and policy relevant data analysis, but it is challenging in some regions due to the commingling of different sectors geographically, and because of limited empirical data on non-oil and gas methane emissions. The empirically based sector disaggregation presented here is a first step. Future work will continue to explore and refine these methods as both still have uncertainties that need to be better assessed.

Gross gas production normalized oil and gas methane loss rates, expressed here as the percentage of methane emitted relative to total methane produced (Eq. 2), are calculated by dividing the total oil and gas methane emissions estimated from MethaneAIR data by the gross methane produced in the measured regions. Gross methane production is estimated using 2023 gross natural gas production data (Enverus, 2024) and basin-specific gas compositions (Table S2) that are consistent with previous literature assumptions on methane composition. 2Methanelossrate(%)=Oilandgasmethaneemissions(kgh-1)(Grossnaturalgasproduction(kgh-1)×methanecontent)×100 As an additional metric for comparisons, we also computed energy-normalized methane intensities (in kg CH₄ GJ⁻¹) by dividing oil and gas methane emissions estimated from MethaneAIR by the combined gross oil and gas production (Enverus, 2024) in the measured regions, similar to intensities reported by the International Energy Agency for their annual Global Methane Tracker (IEA, 2025). See Sect. S5 for additional discussion and comparisons of both metrics across all measured basins.

We spatially aggregated overlapping MethaneAIR flights in the 12 measured oil and gas basins to produce an estimate of total methane emissions and associated uncertainties for each basin. Two separate approaches for aggregation were explored, which we defined as (1) unique overflown area (UOA) averaging, and (2) area-normalized averaging, with both methods producing similar results (Fig. S4). Results from the UOA averaging method are presented in the main text, and comparisons and additional discussion on the other method can be found in Sect. S3.

For the UOA averaging method, we first mapped the spatial domains of each MethaneAIR flight to identify areas that were uniquely overflown by the same combination of flights, which creates a subset of smaller spatial domains (Fig. S3). Next, we iterated through the subset of smaller spatial domains (i.e., denoted as UOA, or unique overflown areas) and averaged both the point source and area emissions quantified from the corresponding flights. The resulting averages of all UOAs are then summed to produce a total estimate of methane emissions for the aggregated flight domains at the basin-level.

To calculate basin-level uncertainties in the dispersed area emissions using the UOA approach, we first adjust the uncertainties for the dispersed emissions for each UOA based on the percentage of area covered using Eq. (3), where Uf is the uncertainty in the dispersed area emissions at the flight level, Af is the area covered by the entire flight, and AUOA is the area covered by the UOA which is a subset of the entire flight domain. This adjustment accounts for the inherently higher uncertainties contained within spatial subsets of the entire flight domain, with the assumption that uncertainties are uniform across the domain. Additional refinement to the uncertainties in the dispersed area emissions should incorporate parameters such as the effects of albedo, terrain, and weather conditions to produce more accurate estimates of the uncertainties across different portions of the spatial domain for a given flight. A comparison of basin-level uncertainties with and without the area-based adjustment show very minimal differences (1 %–3 %), except in the Permian and Denver-Julesburg basins where the area-adjusted approach increases the uncertainties by > 2 × due to more unique flights occurring in these basins. That said, the area-adjusted approach is possibly a more conservative method for estimating uncertainties in these two basins. 3U=AfAUOAUf Next, using the adjusted dispersed area emissions uncertainties (Eq. 3) we propagate the uncertainties for all flights corresponding to a UOA (Eq. 4) where Ui is the area uncertainty from flight i for a UOA, n is the total number of flights within the UOA, and UUOA is the uncertainty for the UOA. 4UUOA=Ui2+…+Un2n To calculate the uncertainties of point sources for each UOA, we resampled point sources (n = 5000) from flights within a UOA assuming the uncertainty for a single point source follows a normal distribution with parameters of the distribution as the point source standard deviation and the mean as the quantified emission rate. The 2.5th and 97.5th percentiles from the resulting distribution are used as the associated uncertainties. Then, we used Eq. (4) to determine the average point source uncertainties for each UOA.

The overall uncertainties for aggregated basin-level total emissions are estimated using Eq. (5) for the area emissions and point source emissions separately, where Utotal is the aggregated basin-level uncertainty, Ui is the percentage uncertainty for an emission estimate from a UOA within the basin, and xi is the associated emission rate for the UOA. Finally, Eq. (5) is used again to combine point source and area emissions uncertainties to produce the overall uncertainty bounds for the aggregated basin-level and national-level total emissions estimates. Uncertainties on the basin-level oil and gas estimates and associated loss rates incorporate the quantification uncertainty as described above (for both area and point source quantification), and an estimate of uncertainty related to the subtraction of literature-based estimates of non-oil and gas emissions (Eq. 1) using the standard deviation of the bootstrapped distribution. 5Utotal=Ui⋅xi2+…+Un⋅xn2xi+…+xn It is important to note that the resulting estimates represent the total methane emissions for the area within each basin that was explicitly measured by MethaneAIR across multiple flights, and not the entire geographic extent of each basin (Fig. 1). However, these measured areas cover more than two-thirds of each basin's total oil and gas production with several exceeding 90 % of the basin's production (except for the Bakken and Greater Green River basins) and combined make up 70 % of the CONUS onshore oil and gas production in 2023 (Table 3).

Figure 2 and Table 3 show total methane emissions estimated by aggregating MethaneAIR data collected in 12 major oil and gas producing basins, delineated by estimated relative contributions of oil and gas and non-oil and gas sources. When considering methane emissions from all sectors, the Permian, Appalachian, and Haynesville-Bossier basins rank highest in terms of absolute methane emissions. In addition to oil and gas methane emissions, the Appalachian has significant emissions from the coal and waste sectors, which was also observed in previous work (Barkley et al., 2019; Cusworth et al., 2022). We estimate that around 14 % of emissions in the Haynesville-Bossier basin are from non-oil and gas methane emissions, most of which are likely from the waste sector (Maasakkers et al., 2023), and for the Permian, we found that almost all (> 95 %) emissions are from oil and gas activity (Fig. 2).

The Barnett and Denver-Julesburg basins are additional regions with larger contributions of non-oil and gas methane emissions (> 35 %), which we attribute primarily to the agriculture and waste sectors (Crippa et al., 2024; Cusworth et al., 2022; Lu et al., 2023; Maasakkers et al., 2023; Peischl et al., 2018). However, based on our analysis, we estimate that oil and gas methane emissions make up the majority of methane emissions in all 12 basins, ranging from 57 %–99 % of the total. As discussed in Sect. 2.3.2, these percent contributions have varying levels of uncertainty due to limited data.

It is important to note that estimates for some basins are based on a single MethaneAIR flight (Table 3), whereas others are based on several flights occurring over the span of weeks or months. Repeat overpasses throughout the year are needed to produce a more representative estimate of basin-level methane emissions.

Figure 3 and Table 3 show the estimated total oil and gas methane emissions and gas normalized loss rates over individual oil/gas basins (following methods discussed in Sect. 2.3.2). When considering only oil and gas related methane emissions, the Permian, Appalachian, and Haynesville-Bossier remain as the highest emitting basins. These basins are also the top oil and gas producers out of all basins included in this study (Table 1). The Permian is dominated by oil production but has significant associated gas production, accompanied by increasing new oil and gas development and high amounts of flaring, all of which could lead to higher observed methane emissions (Lyon et al., 2021; Varon et al., 2023; Zhang et al., 2020).

While absolute emissions reveal important insights as noted above, methane loss rates are an important metric to consider when comparing methane performance across basins with variable levels of gas production. For instance, gas-dominant basins with high well-site productivity (Appalachian, Haynesville-Bassier) have the lowest methane loss rates (< 1 %), despite having some of the highest absolute emissions (Fig. 3, Table 3). Oil-dominant or mixed oil/gas basins (e.g., Permian, Greater Green River, Eagle Ford, Bakken) tend to have higher methane loss rates (2 %–5 %). Relatively mature basins where oil and gas production and well site infrastructure is dominated by large populations of aging, low producing wells such as the Piceance and Uinta have the highest observed methane loss rates (> 7 %) (Fig. 3, Table 3), likely due to fugitive methane emissions that continue to occur even as production declines (Lin et al., 2021; Omara et al., 2022). We also estimated energy-normalized methane intensities (kg CH₄ GJ⁻¹) for each basin to compare oil and gas methane emitted relative to each basin's combined oil and gas production, which are discussed in Sect. S5.

Figure 4A shows basin-level estimates of total methane emissions quantified by MethaneAIR compared to total emissions reported by the EPA (Maasakkers et al., 2023) for the same domains. Note that EPA estimates are for 2020 (the most recent year available at the time of writing), and MethaneAIR measurements were collected in 2023. Depending on the basin, MethaneAIR estimates of total methane emissions range from 1.8 (Barnett) to 8.2 (Greater Green River) times higher than EPA estimates. While it is possible that some of these differences may be due to actual changes in emissions between 2020 and 2023 (e.g., from changes in activity), it is unlikely that such changes would result in the large discrepancies observed, suggesting that underreporting of emissions remains an issue for these regions.

Basin-level oil and gas methane emissions estimated by MethaneAIR similarly range from 1.3 (Barnett) to 7.9 (Greater Green River) times higher than EPA estimates (Fig. 4B), suggesting that observed discrepancies can be primarily attributed to the oil and gas sector. Despite several previous measurement studies finding similar differences between measured and reported emissions (Alvarez et al., 2018; Omara et al., 2024; Sherwin et al., 2024; Zhang et al., 2020), some of which date back to over a decade ago (Brandt et al., 2014), our analysis indicates that underreporting continues to be prevalent for major oil and gas producing basins in the US, which must be addressed if such inventories are to be used to inform mitigation and track promised reductions over time.

To better contextualize this comparison, we assess the relative contributions of EPA's reported oil and gas methane emissions from the MethaneAIR measurement domains compared to their reported emissions for the rest of the onshore CONUS. EPA estimates only 21 % (1.6 Tg yr⁻¹, 0.4 % loss rate) of the total CONUS onshore oil and gas methane emissions are from sources within the MethaneAIR domains, which are responsible for more than 70 % of total onshore oil and gas production in the CONUS for 2023. This suggests that EPA estimates the other 79 % of the onshore CONUS oil and gas methane emissions (6.1 Tg yr⁻¹, 3.6 % loss rate) are from the regions outside of the MethaneAIR domains, which make up only 30 % of total onshore oil and gas production for the CONUS in 2023. Considering these findings, it is possible that the discrepancies between EPA and MethaneAIR are partly attributable to EPA's use of methane emission factors (e.g., estimate of average methane emitted per gas well) that may be unrepresentative across basins.