Quantifying oil and natural gas system emissions using one million aerial site measurements

doi:10.21203/rs.3.rs-2406848/v1

Aerial methane surveys of oil and natural gas systems have discovered large emissions, which are missing or vastly undercounted in official estimates. We integrate approximately one million aerial site measurements into regional emissions inventories, employing empirically-grounded emissions simulations to estimate small emissions. We infer emissions inventories for six regions in the United States comprising 52% of onshore oil and 29% of gas production over fifteen aerial campaigns. Total estimated emissions range from 9.63% [9.03%, 10.37%] of natural gas production, roughly nine times the US government estimate, to 0.75% [0.65%, 0.85%] in a high-productivity gas-rich region. Aerially measured emissions at 0.05%-1.44% of well sites contribute 50%-81% of total emissions in twelve of fifteen campaigns. The social cost of methane emissions from these measured regions is roughly $9.4 billion per year, in addition to roughly $1 billion in lost sales. This highlights the importance of incorporating comprehensive remote sensing surveys into emissions inventories and efforts to benchmark and reduce methane emissions.

Earth and environmental sciences/Environmental sciences/Environmental impact

Physical sciences/Energy science and technology/Fossil fuels/Natural gas

Physical sciences/Energy science and technology/Energy modelling

Scientific community and society/Energy and society/Energy policy

Methane

hyperspectral imaging

remote sensing

power law

emissions inventory

simulation

super-emitter

Accurately estimating methane emissions from the oil and natural gas sector is vitally important for mitigating climate change. Current global inventories estimate oil and gas sector emissions of 84 Tg(CH₄)/yr [72, 97] ¹. These emissions represent 16% of global anthropogenic greenhouse gas emissions from fossil fuels assuming a 20-year global warming potential ^1–3. However, this estimate, and all estimates like it, rely on coarse emission factors used with often-incomplete data. Many countries simply leverage the most general emission intensities from – for example – the United States Environmental Protection Agency’s Greenhouse Gas Inventory, or GHGI ^4,5. In the last decade, efforts in the scientific literature have scaled actual ground measurements at up to ~1,000 sites to estimate industry-wide emissions, finding that emission factor approaches like the above can undercount actual emissions by as much as 40% ⁶. Even more recently, aircraft-based technologies have enabled a 1,000-fold increase in the size of surveys and have found very large emissions sources. The results of these aerial surveys have not been incorporated into previous systematic inventories because no method existed to combine aerial measurements of large emission sources with inventory methods for smaller sources. In this paper, we close this gap by making a first estimate of large-scale comprehensive emissions across major producing regions by merging results of approximately 1 million site visits with an empirically-grounded statistical model of smaller emission sources. ^6,7.

In the 1990s, countries across the world began producing systematic inventories of emissions of methane and other greenhouse gasses to track progress toward climate goals ⁸. Many national inventories rely on scaling up relatively small datasets of component-level measurements, with underlying empirical data collection often conducted decades ago in the United States ^8,9. Rutherford et al. recently demonstrated that updating the underlying component-level data to include the most recent studies largely reconciles these emission factor methods with the ground measurement literature ⁵.

However, recent remote sensing surveys conducted by airplanes and satellites have discovered a substantial number of point-source methane emissions that are up to three orders of magnitude larger than those reported by ground-based well site methane surveys ^10–18. In at least some cases, these low-probability but high-consequence sources can contribute the majority of methane emissions from an oil and gas producing region ^13,14. Because ground-based surveys remained dominant in the literature throughout the 2010s, these remote sensing results suggest that emission factor methods such as the GHGI produce an even more significant undercount of total methane emissions than previously thought ⁶.

Therefore, accurately characterizing the statistical distribution of methane emissions from the oil and gas system, including the likelihood of low-probability, high-emission events – sometimes termed “super-emitters” – is key to accurately estimating total methane emissions from an oil and gas-producing region ^7,19–22. Ground-based methods are inherently challenged to find these due to the low probability of occurrence at a given site and small sample sizes of ground campaigns.

In order to make regional estimates including data from airplane-based point-source sensing methods, we need to understand emissions distributions both above and below the detection limit of the technology in question. Understanding the full distribution of methane emissions is also necessary when evaluating equivalence between different emission detection and mitigation programs. A theoretical basin dominated by large emissions will benefit from a scalable approach, such as airborne, that can rapidly detect large sources, while another basin with most methane lost to smaller emissions will be more effectively surveyed by ground-based or other high-sensitivity approaches. Note that while airplane-based area flux estimation methods can produce regional estimates that do not attribute emissions to sources ⁶, this study focuses on airplane-based point source methods, which can detect individual sources at the facility level.

Air- or ground-based surveys each necessarily miss some emissions. However, quantifying that missing amount is a challenge. Ground-based surveys can see most sources reliably, although they may miss elevated sources such as tanks and flares ²³, but are slow and expensive and therefore are generally limited to small sample sizes of tens to hundreds of sites. This limited sampling potentially misses large, low-probability emissions. In contrast, aerial surveys can cover vast swaths of landscape, but have a higher minimum detection limit that prevents them from seeing smaller sources. Thus, these two methods each exhibit complementary strengths and weaknesses, but past work has had difficulty reliably to synthesize these disparate results into a complete distribution of methane emission rates. Johnson et al. take steps in this direction, combining aerial measurements with emissions simulation methods for 8% of well sites in British Columbia ²⁴.

In this paper we develop an approach to construct such a complete distribution, from the smallest to the largest sources at the regional scale, leveraging comprehensive aerial surveys as well as a state-of-the-art component-level emission simulation tool ⁵. We define a comprehensive survey as including measurements of at least 50% of well sites and 80% of natural gas production in a region. Such region-specific distributions can inform methane emissions inventory development and guide deployment of methane-sensing technologies whose spatial coverage and detection limits most closely match the needs of a particular area.

A key challenge in combining data from various methods is avoiding errors when datasets are joined. For example, data generated should cover the entire range of expected emission sizes, avoiding gaps at intermediate sizes. Also, one must avoid double-counting emission prevalence where two methods may both be able to see emissions of a given size. Another significant consideration is ensuring the surveyed area has an emissions profile that is representative of the area of interest, e.g. an entire oil and gas-producing basin. We describe below an approach that combines emissions simulation and aerial measurement methods into a single distribution, avoiding double-counting and mitigating data gaps in intermediate size ranges for all assets within a surveyed area.

We generate a unified methane emissions distribution for surveyed oil and gas assets in six US oil and gas producing regions: the Permian, San Joaquin, Denver-Julesburg, Uinta and Fort Worth basins, and Appalachian Pennsylvania (illustrated in Figure 1A). Eleven of these surveys are comprehensive, with the Pennsylvania and Permian 2020 and 2021 campaigns still covering at least 10% of well sites and 39% of natural gas production. These surveys include 959,573 well site measurements as well as a difficult-to-quantify number of measurements of midstream infrastructure, including compressor stations, gas processing plants, and pipelines. The surveyed areas within these regions comprise 29% of onshore US gas production and 52% of oil production. We combine fifteen large aerial surveys, conducted by Kairos Aerospace (Kairos) and researchers leading the Carbon Mapper (CM) project, with an empirically-grounded emissions simulation method introduced in Rutherford et al. 2021. We refine this model with input parameters based on regional characteristics, and estimate small-source midstream emissions based on state-level and national GHGI data ^5,9,13,25. Given proper input data, this emissions simulation method produces results that largely reproduce the distribution of site-level emissions from a synthesis of the ground-based site-level methane measurement literature ^5,20. We demonstrate that our combined emissions distribution is consistent with major ground-based and aerial methane measurement studies.

For all oil and gas production assets covered in each surveyed region, we compute two well site-level emissions inventories: One based on aerial measurements and the other using the Rutherford et al. emissions simulation tool ⁵. We consider a well site to be a point location that may contain multiple wells and supplementary equipment such as liquids tanks, flares, and separators. Figure 1B provides a high-level overview of the 1000-realization Monte Carlo-based method we use to synthesize these two emissions inventories into a unified estimate of the distribution of methane emissions across surveyed well sites, which allows estimation of total emissions.

Aerial surveys also cover midstream assets such as gathering and transmission pipelines, compressor stations, and gas processing plants. We estimate aerially measured and partially detected midstream emissions using the same approach as above. We do not have sufficient asset location data or emissions simulation tools for midstream infrastructure to estimate site-level emissions below the aerial detection limit. We instead estimate these emissions based on the EPA national and state-level GHGI, removing the fraction of emissions underlying those estimates that would be detectable by Kairos or CM ^9,25. All Kairos data used in this study are fully anonymized, and include no identifying information for covered operators or their assets. See Materials and methods for further detail.

Estimation at the site-level, rather than site-visit level, is key. Treating each aerial site visit as a completely independent measurement can introduce significant bias if a subset of assets that is not representative of the full survey area is disproportionately over-sampled, as demonstrated in the Supplementary Information (SI), Section S8. Note that we do not estimate methane emissions from local distribution or oil refining and transportation, which occur at facilities largely outside the surveyed regions.

1.1 Methane loss rates vary widely over space and time

Estimated methane loss rates – the emitted fraction of methane produced from oil and natural gas activity in a given region – vary widely across the studied US regions. Estimated rates are as low as 1.08% [0.98%, 1.18%] in the Denver-Julesburg basin in Colorado in 2021, and 0.75% [0.65%, 0.85%] in a high-productivity area of the Pennsylvania portion of the Appalachian basin, as shown in Figure 2. In contrast, for the New Mexico Permian 2018-2020 campaign, the loss rate is 9.63% [9.03%, 10.37%], an order of magnitude higher. The remaining campaigns range from roughly 2% to 6% methane loss rates. The production-weighted loss rate across all fifteen campaigns is 2.97% [2.78%, 3.18%], rising to 4.77% [4.51%, 5.07%] excluding the four campaigns focusing on high-productivity sub-regions. These loss rates assume a conservatively high methane fraction of 90% from ⁶. If the actual methane fraction is lower, these loss rates would increase correspondingly, as discussed in the SI, Section S7.

Most of these estimates are far larger than the national EPA GHGI, which places the 2020 US-wide onshore methane loss rate at 1.01% [0.81%, 1.22%], after excluding municipal distribution systems, crude oil transportation and refining, and post-meter emissions for consistency with this study ⁹. These loss rates also generally exceed state-specific EPA inventories, shown as dotted lines in Figure 2, although these estimates roughly align in the Denver-Julesburg and the studied region of Pennsylvania, and the EPA value actually exceeds our estimate for four of the five San Joaquin campaigns ²⁵. See the Materials and methods for further description of our treatment of EPA GHGI estimates.

Multiple surveys across the oil-focused San Joaquin and Permian basins demonstrate substantial variation in loss rates over time. The five San Joaquin campaigns find loss rates as low as 2.52% [2.20%, 2.87%] in fall 2021, and as high as 5.56% [4.97%, 6.23%] in 2017. All San Joaquin campaigns cover at least 80% of the basin’s gas production and over 80% of all actively producing well sites. Across the Permian 2020 and 2021 campaigns, which cover similar areas, the methane loss rate varies from 2.10% [1.94%, 2.27%] to 2.81% [2.59%, 3.05%].

Even large surveys of the same region can produce divergent results if they cover different areas. Loss rates in the five Permian campaigns vary from 2.10% [1.94%, 2.27%] in the fall 2021 campaign, focusing on high-productivity areas, to 9.63% [9.03%, 10.37%] in the New Mexico Permian from 2018-2020. The largely overlapping 2019 survey of both Texas and New Mexico finds 5.29% [5.07%, 5.51%]. See the SI, Sections S5 and S12 for further detail.

This area-specific variation highlights the need to use comprehensive, or at least representative, aerial surveys when estimating regional emissions. For this reason, Figure 2 uses semi-transparent bars to represent methane loss rate estimates from the Permian 2020 and 2021 campaigns, as well as the Pennsylvania 2021 campaign, all of which disproportionately focus on high-productivity areas and cover less than 80% of natural gas production and less than 50% of well sites in the region in question. For simulated well site emissions, we account for the productivity of surveyed well sites, as described in the SI, Section S1.4 and S13, but this simply improves the fidelity of simulated emissions within the covered region. As a result, our estimates from these less comprehensive campaigns should not be extrapolated to the full region (e.g. the entire Permian basin or all of Pennsylvania, respectively).

In all cases, midstream emissions are a significant fraction of the total. Midstream emissions represent 45%-57% of total estimated oil and gas emissions in the Permian basin. This falls to 31%-55% in the San Joaquin basin and 18% in the Uinta. We estimate midstream emissions at 32-37% of the total in the Denver-Julesburg basin, although aerially measured midstream emissions are only 3-10% of the total, suggesting that simulated midstream emissions may be an overestimate in this case. See Materials and methods for further discussion of simulated midstream emissions, which are derived from the national and state-level GHGI reports ^9,25.

Note that campaigns in the same region with comparable well site coverage may cover different amounts of midstream infrastructure, which may affect estimated midstream emissions estimates. We do not have sufficient midstream asset location data to quantify this effect here. See the SI, Section S12 for coverage information for each campaign.

Aerially measured emissions play a major role in nearly all basins, contributing 50-81% of the total in all Permian, San Joaquin, Pennsylvania, and Fort Worth campaigns. This rises as high as 84% in the New Mexico Permian after accounting for missed emissions in the partial detection range of the aerial system. The fraction of aerially measured emissions falls to 41% the Uinta, and 14-20% in the Denver-Julesburg. Note that this does not include emissions below the transition point from simulated to aerially measured emissions, which excludes some emissions in the partial detection range if simulated emissions are larger. This means that the fraction of emissions that is aerially detectable is even greater, as illustrated in the SI, Section S3.

Many new methane-sensing technologies have emerged in the past several years. To assess the fraction of total emissions that a technology with a given sensitivity would see in each region, we show the full well site-level emissions distribution for all actively producing well sites covered in each of the fifteen campaigns in Figure 3. Note that these distributions do not include midstream emissions, as we do not have site-level simulated emissions estimates in that case. Each point on these distributions represents a well site emission magnitude, on the x-axis, and the fraction of total estimated regional emissions coming from well sites emitting at least that amount on the y-axis.

Despite their substantial contribution to the total in all cases, aerially detected emissions are present at only a small fraction of sites at a given time. In the Permian basin, an average of 0.86%-1.44% of total well sites are emitting in a given Monte Carlo simulation. This fraction falls to 0.05%-0.09% of sites in the San Joaquin, with the remaining regions between the two. Thus, while previous literature focused on ground-based measurements found that 5% of measurements often contributed 50% of total emissions ²¹, this study finds that less than 1% of measurements contribute over 50% of total emissions in 12 of 15 cases. See the SI, Section S2 for further discussion of the shape of the emissions distribution, which generally resembles a lognormal for simulated emissions and a power law for aerially measured emissions.

For Kairos datasets, we use single-blind controlled release test results to correct for missed emissions in the partial detection range (teal in Figure 2 and Figure 3). If the probability of detecting an emission of a certain magnitude is 1/3 and such an emission is detected, this implies that two emissions of comparable magnitude were likely missed by the survey. Thus, we account for this in the cumulative emissions distribution by multiplying the contribution of this emission to the total by 3, as described in the SI, Section S1.7. Note that due to the required use of wind reanalysis data, the high end of the partial detection range for Kairos is larger than observed when using in situ field wind measurements, as in ²⁶.

The point at which each distribution transitions from simulated to aerially measured emissions is often at a higher rate than the smallest aerially measured emission. We set this transition point as the site-level emission magnitude beyond which aerially measured emissions are larger than simulated emissions (described further in the SI, Sections S1.8 and S3). As a result, some small emissions detected in most campaigns are not included in the combined distribution. This is likely because they fall within the partial detection range of the system, which can vary depending on the sensitivity of the sensor, flight altitude, automated and manual quality control processes, and local environmental conditions such as wind, sun angle, surface reflectance, and vegetation cover. While we correct for missed emissions in the partial detection range for Kairos surveys using the method from ¹³, commensurate single-blind controlled release testing data to do so for Carbon Mapper. In addition, Carbon Mapper conducted flights at different altitudes across campaigns, as described in the SI, Section S12, affecting the lower detection range.

The absence of a correction in the partial detection range for Carbon Mapper campaigns, coupled with variation in detection capabilities introduced by flight altitude and other above-mentioned factors, likely explains the somewhat higher transition points observed compared to Kairos campaigns. In the San Joaquin and Pennsylvania distributions, relatively flat areas indicate a gap between the largest simulated emissions and the smallest measured emissions, suggesting our estimates of total emissions in these regions may be conservative, due in part to missing emissions in this middle size range that exist but were not captured by our method. Carbon Mapper personnel believe that the gap in Pennsylvania is partially due in part to high vegetative cover. In probability density function form, shown in the SI, Section S2, this gap between simulated and measured emissions appears as a local minimum in emission frequency, generally ranging between 10 kg/hr and 100 kg/hr, followed by a local maximum for aerially measured emissions. It is unclear whether the underlying distribution has a true local minimum in this size range in some regions, or whether this is entirely an artifact introduced by sensor minimum capabilities.

In some cases, aerially detectable emissions significantly exceed the aerially measured portion of the combined distribution. The vertical lines in Figure 3 represent the minimum emission detected by each aerial technology in each region. In some instances, this is close to the transition point, as in the Kairos New Mexico Permian campaign, which has a minimum detected emission of 7 kg/hr and a transition point of 19 kg/hr. This indicates that while aerially measured emissions contribute 83% of total emissions, Kairos-detectable emissions constitute 88%. The gap is larger for the Denver-Julesburg, where aerially measured emissions are 14% of the total and Carbon Mapper-detectable emissions are 45% of the total. While some of these detectable emissions may lie in an aerial technology’s partial detection range, this illustrates that the distributions in this paper provide a conservative estimate of the fraction of emissions these technologies will see when deployed in the field.

Regions with high emission rates are not necessarily dominated by aerially measured emissions. In the Uinta, aerially measured emissions comprise a relatively small fraction of total production site emissions, 41%, even though the overall methane loss rate is the second highest of all campaigns. In this case, production site simulated emissions alone constitute a 3.00% [2.86%, 3.13%] methane fractional loss rate, including only emissions below the transition point of 52 kg/hr. A major reason for these large simulated emissions in the Uinta basin is the widespread use of emission-prone gas-driven pneumatic controllers, shown in more detail in the SI, Sections S1.4.2 and S4. This illustrates the importance of capturing regional variability in well site composition when simulating emissions. See the SI, Section S3 for full aerial and simulated distributions for each campaign.

The surveyed regions in this study contribute an estimated 6.3 million t/yr of methane emissions, equivalent to total greenhouse gas emissions from France. At $3/mcf (~1¢/kWh) this amounts to $1.1 [$1.0, $1.2] billion in annual lost gas sales. The environmental damage from these emissions is roughly five times larger, at $9.4 [$8.9, $10.0] billion assuming a $1,500/t social cost of methane, discussed further in the SI, Section S10 ²⁷.

Our study includes 959,573 well site measurements and also captures substantial associated midstream infrastructure, representing a three-order-of-magnitude advance over the ground-based measurement literature. A 2018 synthesis of nearly all previous ground-based well site-level methane measurements included just over 1,000 site measurements across multiple studies performed over roughly 5-years ²⁰.

This scale allows us to accurately characterize the contribution of low-probability, high-consequence emissions. We demonstrate through comprehensive aerial surveys and empirically grounded emissions simulation methods, that methane emissions vary widely across regions and over time. Moreover, across all campaigns aerially detectable emissions occur at a small fraction of sites, 1.44% or less, but make an outsized contribution to total regional emissions. This proportion is beyond what would be expected if total estimates were based solely on a bottom-up emissions simulation method.

Previous, much smaller ground-based studies likely under-sampled the largest emitters. These methods likely underestimated further due to systematic difficulties in the primary ground-based methods used to quantify large emissions from high source locations (e.g. a vent stack) or with substantial upward velocity ²⁸. These factors likely contributed to a significantly lower estimate of the methane loss rate, 2.25% [1.88%, 2.79%] for US well sites and midstream infrastructure, than we find in most regions in this study ⁶. See the SI, Section S10 for further detail on possible reasons for underestimation in past literature.

These study results confirm previous findings that simply scaling up component-level measurements, as in the EPA GHGI ⁴, will often result in consequential underestimation of basin-scale emissions ⁶. In all campaigns, we quantified significant methane loss from large, aerially-detected sources which were vastly underrepresented or completely absent in prior datasets, as shown in the SI, Section S3. That said, our results match closely with the state-level GHGI in the Denver-Julesburg, San Joaquin, and a high-productivity region of Pennsylvania, suggesting that emissions simulation methods can produce accurate estimates in some contexts.

Illuminating the full extent of undercounting that may be occurring in official inventory estimates across the entire US and internationally will require additional comprehensive measurement campaigns. Given the regional variability in emission rates we show in this work, an accurate estimate of US emissions will require surveying a large fraction of the 48% of onshore oil and 71% of natural gas production not covered here. Correspondingly, the working assumption for all global oil and natural gas-producing regions should be that aerially detectable sources contribute significantly to the overall emissions total until proven otherwise through comprehensive measurement campaigns.

This work also points to the need for better fidelity in the data underlying emission simulation methods. Many of the original component-level measurements that underly the GHGI were conducted in the 1990s, before the advent of hydraulic fracturing and numerous other technical developments ²⁹. In addition, both the GHGI and our own emissions simulation method, adjust regional equipment counts but do not account for regional variation in component-level emission incidence and severity ^5,9, which may in fact very regionally. Furthermore, many countries rely on EPA emission factors when reporting methane emissions from oil and natural gas systems to the United Nations ⁸. Thus, our findings imply that the resulting estimates of these emissions are most likely conservative for many countries, highlighting the need to complement comprehensive remote sensing surveys with component-level emission measurement campaigns across the world’s oil and gas-producing regions.

We find substantial variability in emissions from the Permian and San Joaquin, even within the same year. In addition, midstream assets are responsible for roughly half of total emissions in many regions. Thus, repeated measurements at regular time intervals of both production and midstream assets, including gathering and transmission pipelines, will be essential for industry and regulators to track progress in reducing methane emissions.

The methane loss rate is highly sensitive to the area surveyed. The New Mexico Permian 2018-2020 campaign yields a methane loss rate almost twice that of the comprehensive Permian 2019 campaign. This highlights the need for measurement-based emissions inventories to rely on comprehensive, or at least large representative surveys of the region in question. Thus, although our surveys collectively cover roughly 29% of 2021 US onshore natural gas production, we do not attempt to estimate a national methane loss rate, which would require significant extrapolation outside the surveyed areas. Generating accurate national and region-specific methane loss rates will also require further improved estimates of regional variation in the methane fraction of natural gas produced at well sites.

The methane loss rates estimated in this paper will likely be of interest to the life cycle assessment community for estimating the greenhouse gas impact of industries and activities that use oil and natural gas. Note that these loss rates represent the fraction of total methane produced that is emitted from oil and gas well sites and midstream infrastructure. For the purposes of life cycle assessment, some fraction of the emitted methane should be allocated to the oil produced, with the remainder allocated to natural gas. See the SI, Section S9 for further discussion.

The Denver-Julesburg data suggest that it may be possible to limit aerially detectable emissions from production and midstream to a level roughly consistent with component-based simulations. Contributing factors to this lower emission rate likely include management practices, the regulatory environment, and system design, as well as local geology.

This study hints at significant methane emissions reductions that could be achieved by highly targeted intervention approaches. This could be accomplished by focusing on the small fraction of sites that contribute a large proportion of overall emissions, and modifying equipment or operations at those sites to reduce emissions. These sites can be effectively flagged across large geographic areas by aircraft or dedicated satellites, such as those characterized in ³⁰. Future emissions modeling to assess the efficacy of such rapid screening methods in the six regions characterized here should use the distributions presented in this paper, or similar methods, at least for the campaigns with comprehensive spatial coverage. Furthermore, updated distributions can be fed into existing, peer-reviewed numerical models to generate more accurate field emissions and assess basin-specific efficacy of different leak detection and repair technologies and deployment strategies ^31,32.

The well site-level emission size distributions produced in this paper, as well as the aerially measured emissions distributions for midstream assets, provide a powerful tool for key stakeholders. Environmental regulators, companies, and intergovernmental agencies now have a firm empirical basis upon which to assess the utility of the expanding variety of methane sensing technologies across different regions and use cases. Without incorporating comprehensive measurement campaigns, existing national and regional emissions inventories can easily be low by a factor of two or more. Tracking progress requires repeated surveys at regular intervals.

A substantial fleet of dedicated point-source methane satellites has already begun deployment, and soon may enable measurement with meaningful sensitivity anywhere on Earth. This will represent an enormous advance for understanding emissions in every hydrocarbon-producing country. Translating these data, once available, into usable emissions inventories will require several advances, including a method that combines measurements across scales. The method introduced in this study forms the basis for doing precisely that.

Remote sensing allows us to track methane emissions with unprecedented accuracy, speed, and scale. This forms a powerful basis for rapid, unambiguous reductions in global methane emissions.

Code availability

The data and code required to reproduce the key results of this article, as well as 100,000 random samples from each simulated emissions distribution in this study, are available at https://github.com/esherwin/MethaneDistributions.

Data availability

The remaining Kairos Aerospace data from this study are not available for open release due to confidentiality concerns, Kairos Aerospace is committed to working with research groups studying methane emissions. Access may be granted, but must be done directly through Kairos Aerospace. Interested researchers should contact research-collaborations@ kairosaerospace.com. All methane data from airborne campaigns led by the Carbon Mapper/JPL team since 2016 are available at https://doi.org/10.3334/ORNLDAAC/1727, https://doi.org/10.5281/zenodo.5610307, https://doi.org/10.5281/zenodo.5606120, http://www.nature.com/articles/s41586-019-1720-3, https://pubs.acs.org/doi/10.1021/acs.estlett.1c00173.

Funding sources

This study was funded by the Stanford Natural Gas Initiative, an industry consortium that supports independent research at Stanford University. Funding for AVIRIS-NG and GAO flight operations and/or data analysis referenced in this paper was provided by NASA’s Carbon Monitoring System and Advanced Information System Technology programs as well as Carbon Mapper, RMI, the Environmental Defense Fund, the California Air Resources Board, the University of Arizona, and the US Climate Alliance. Funding for Colorado overflights was provided by the Mark Martinez and Joey Irwin Memorial Public Projects Fund with the support of the Colorado Oil and Gas Conservation Commission and the Colorado Department of Public Health and Environment. Portions of this research were carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). The Global Airborne Observatory (GAO) is managed by the Center for Global Discovery and Conservation Science at Arizona State University. The GAO is made possible by support from private foundations, philanthropic individuals, and Arizona State University.

Acknowledgments

J. Kruguer of Kairos Aerospace provided data management support. T. Lauvaux, C. Giron and A. d’Aspremont provided satellite-derived data to support analysis of power law behavior. This work benefitted from discussions with N. Boness, S. El Abbadi, B. Hmiel, E. Kort, and R. Jackson.

Author contributions

Conceptualization: E.D.S., A.R.B., E.S.F.B., B.B.J.; Resources: A.R.B., E.S.F.B., R.M.D.; Data curation: E.D.S., E.S.F.B., E.B.W., P.V.Y., D.H.C., Z.Z., J.S.R., Y.C.; Software: E.D.S., J.S.R., Z.Z., Y.C.; Formal analysis: E.D.S., J.S.R., Z.Z.; Supervision: A.R.B; Funding acquisition: A.R.B., E.S.F.B., R.M.D.; Investigation: E.D.S., E.S.F.B., E.B.W., D.H.C., A.K.T., A.K.A.; Visualization: E.D.S., Z.Z.; Methodology: E.D.S., J.S.R., E.B.W., P.V.Y.; Writing – original draft: E.D.S.; Project administration: E.D.S., A.R.B.; Writing – reviewing and editing: All authors.

Competing interests

The authors declare the following competing financial interest(s): E.S.F.B., P.V.Y., B.B.J., and E.B.W. are employees of Kairos Aerospace. RM.D., D.H.C., and A.A. are employees of the University of Arizona and seconded to the non-profit organization Carbon Mapper. Remaining authors have no competing interests to declare.

Supplementary Information is available for this paper.

Saunois, M. et al. The Global Methane Budget 2000–2017. Earth Syst. Sci. Data 12, 1561–1623 (2020).
Friedlingstein, P. et al. Global Carbon Budget 2020. Earth Syst. Sci. Data 12, 3269–3340 (2020).
EPA. Greenhouse Gas Emissions: Understanding Global Warming Potentials. https://www.epa.gov/ghgemissions/understanding-global-warming-potentials (2017).
US EPA. Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2017. https://www.epa.gov/sites/production/files/2019-04/documents/us-ghg-inventory-2019-main-text.pdf (2019).
Rutherford, J. S. et al. Closing the methane gap in US oil and natural gas production emissions inventories. Nat Commun 12, 4715 (2021).
Alvarez, R. A. et al. Assessment of methane emissions from the U.S. oil and gas supply chain. Science eaar7204 (2018) doi:10.1126/science.aar7204.
Brandt, A. R. et al. Methane Leaks from North American Natural Gas Systems. Science 343, 733–735 (2014).
Carras, J. N. et al. 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Chapter 4 - Fugitive Emissions. https://www.ipcc-nggip.iges.or.jp/public/2006gl/pdf/2_Volume2/V2_4_Ch4_Fugitive_Emissions.pdf (2006).
EPA. Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2020. U.S. Environmental Protection Agency. https://www.epa.gov/ghgemissions/inventory-us-greenhouse-gas-emissions-and-sinks-1990-2020 (2022).
Duren, R. M. et al. California’s methane super-emitters. Nature 575, 180–184 (2019).
Irakulis-Loitxate, I. et al. Satellite-based survey of extreme methane emissions in the Permian basin. Sci. Adv. 7, eabf4507 (2021).
Cusworth, D. H. et al. Intermittency of Large Methane Emitters in the Permian Basin. Environ. Sci. Technol. Lett. 8, 567–573 (2021).
Chen, Y. et al. Quantifying Regional Methane Emissions in the New Mexico Permian Basin with a Comprehensive Aerial Survey. Environ. Sci. Technol. 56, 4317–4323 (2022).
Cusworth, D. H. et al. Strong methane point sources contribute a disproportionate fraction of total emissions across multiple basins in the United States. Proc. Natl. Acad. Sci. U.S.A. 119, e2202338119 (2022).
Frankenberg, C. et al. Airborne methane remote measurements reveal heavy-tail flux distribution in Four Corners region. Proc Natl Acad Sci USA 113, 9734–9739 (2016).
Varon, D. J. et al. Satellite Discovery of Anomalously Large Methane Point Sources From Oil/Gas Production. Geophys. Res. Lett. 46, 13507–13516 (2019).
Varon, D. J. et al. High-frequency monitoring of anomalous methane point sources with multispectral Sentinel-2 satellite observations. Atmos. Meas. Tech. 14, 2771–2785 (2021).
Lauvaux, T. et al. Global Assessment of Oil and Gas Methane Ultra-Emitters. Science 375, 557–561 (2022).
Zavala-Araiza, D. et al. Super-emitters in natural gas infrastructure are caused by abnormal process conditions. Nat Commun 8, 14012 (2017).
Omara, M. et al. Methane Emissions from Natural Gas Production Sites in the United States: Data Synthesis and National Estimate. Environ. Sci. Technol. 52, 12915–12925 (2018).
Brandt, A. R., Heath, G. A. & Cooley, D. Methane Leaks from Natural Gas Systems Follow Extreme Distributions. Environ. Sci. Technol. 50, 12512–12520 (2016).
Weller, Z. D., Hamburg, S. P. & von Fischer, J. C. A National Estimate of Methane Leakage from Pipeline Mains in Natural Gas Local Distribution Systems. Environ. Sci. Technol. 54, 8958–8967 (2020).
Tyner, D. R. & Johnson, M. R. Where the Methane Is—Insights from Novel Airborne LiDAR Measurements Combined with Ground Survey Data. Environ. Sci. Technol. 55, 9773–9783 (2021).
Johnson, M., Conrad, B. & Tyner, D. Creating Measurement-Based Oil and Gas Sector Methane Inventories using Source-Resolved Aerial Surveys. https://www.researchsquare.com/article/rs-2203868/v1 (2022) doi:10.21203/rs.3.rs-2203868/v1.
EPA. State GHG Emissions and Removals. https://www.epa.gov/ghgemissions/state-ghg-emissions-and-removals (2022).
Sherwin, E. D., Chen, Y., Ravikumar, A. P. & Brandt, A. R. Single-blind test of airplane-based hyperspectral methane detection via controlled releases. Elementa: Science of the Anthropocene 9, 00063 (2021).
Yarmuth, J. A. Inflation Reduction Act of 2022. Sec. 60113, Methane Emissions Reduction Program. Public Law vols 117–169 (2022).
Edie, R. et al. Constraining the accuracy of flux estimates using OTM 33A. Atmos. Meas. Tech. 13, 341–353 (2020).
EPA. Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2020. https://www.epa.gov/system/files/documents/2022-04/us-ghg-inventory-2022-main-text.pdf (2022).
Sherwin, E. D. et al. Single-blind validation of space-based point-source methane emissions detection and quantification. Preprint at https://doi.org/10.31223/X5DH09 (2022).
Fox, T. A., Gao, M., Barchyn, T. E., Jamin, Y. L. & Hugenholtz, C. H. An agent-based model for estimating emissions reduction equivalence among leak detection and repair programs. Journal of Cleaner Production 282, 125237 (2021).
Kemp, C. & Brandt, A. R. Fugitive Emissions Abatement Simulation Testbed FEAST v1.0. 80.

We estimate the full distribution of the magnitude of methane emissions for 15 large-scale aerial surveys of at least 10% of well sites and at least 35% of natural gas production in each of six regions, although the Pennsylvania survey covers only 8% of statewide oil production. This includes campaigns by Kairos in the New Mexico Permian basin and the Fort Worth basin in Texas (focusing on the Barnett shale), alongside campaigns conducted by the Carbon Mapper-led team (including scientists from JPL, the University of Arizona, and Arizona State University) in the Permian basin in New Mexico and Texas (5 campaigns), California’s San Joaquin basin (5 campaigns), the Denver-Julesburg basin (2 campaigns), as well as the Uinta basin and a high-productivity portion of the Appalachian basin in Pennsylvania (1 campaign each) ^{10,12,33–35}.

All campaigns use hyperspectral infrared spectroscopy to detect and quantify methane emissions using the spectral signature of methane in reflected sunlight. The quantification accuracy and minimum detection capabilities of the Kairos technology was independently validated in single-blind controlled release testing in ²⁶. See ³⁶ for further detail surrounding the technology. The Carbon Mapper campaigns were conducted with the Airborne Visible-Infrared Imaging Spectrometer - Next Generation (AVIRIS-NG) spectrometer on a JPL-contract King Air B200 aircraft and an identical very short wavelength infrared (VSWIR) imaging spectrometer on the Global Airborne Observatory (GAO) operated by Arizona State University, both described in ¹². The AVIRIS-NG and GAO systems have also undergone non-blinded controlled release testing to assess minimum detection limits and quantification accuracy ³⁷.

Both teams use data from imaging spectrometers to estimate methane flux rates based on measured atmospheric methane enhancements, retrieved from spectral radiances, combined with estimates of 10 m wind speeds from reanalysis products. For Kairos Aerospace, we combine reported wind-normalized emission rates with HRRR hourly instantaneous wind speed estimates, as in ¹³. Carbon Mapper uses the average HRRR from the nearest nine reported grid values, averaged over the hour before, hour after, and hour of a given measurement, as described in ¹⁰.

Kairos flights were conducted at roughly 900 meters above ground level. Carbon Mapper flights range from 3,000 meters to 8,500 meters, described in detail in Table 18.

Below we describe the steps to construct a complete emissions distribution from a comprehensive aerial measurement campaign:

Step 1: Conduct a comprehensive aerial measurement campaign

To produce an aerial measurement-based regional emissions inventory for oil and natural gas production and midstream activity, one must first conduct a comprehensive aerial survey of the region in question. In this study, we term a survey “comprehensive” if it covers at least 50% of all active oil and natural gas well sites and at least 80% of natural gas production in the region in question, generally an oil and natural gas-producing basin. While future studies may use alternative definitions of “comprehensive”, it is noteworthy that measurement campaigns that focus only on high-productivity or low-productivity areas of a region can produce misleading estimates of the overall regional methane loss rate, as illustrated in the SI, Section S5. See Table 17 and Table 19 for coverage information for each survey in this study.

Step 2: Estimate regional aerially measured emissions using Monte Carlo analysis

We first estimate the distribution of measured emissions in each aerially surveyed region as a function of emission size, using each emission source as the unit of analysis. Each oil or natural gas well site is a potential emission source. In midstream, facilities such as compressor stations and gas-processing plants are potential emission sources, as are pipelines. For pipelines, each detected emission location is considered an emission source.

In many instances, an emission source was surveyed multiple times, with emissions detected during only a fraction of aerial measurements. To account for this, we apply Monte Carlo simulation to characterize the emission profile of the surveyed region. We simulate emissions from each emission source with at least one detected emission, drawing randomly from all aerial measurements at that location, including those with no detected emissions. We then randomly insert simulated error into each quantified emission, based on estimates of quantification uncertainty, discussed further in the SI, Section S1.1. We repeat this stochastic process for 1000 Monte Carlo realizations to capture uncertainty. This method yields an unbiased estimate of total well site emissions in the surveyed region, as described in Chen, Sherwin et al. 2022 ¹³. By analogous logic, it also yields an unbiased estimate of the size distribution of aerially visible emissions, but not the variance of total emissions. See the SI, Section S12.2 for further detail. The resulting emissions inventory covers only aerially detected emissions, treating emissions as zero at all sites at which emissions were not detected.

In the Kairos Fort Worth survey, 8.5% of detected emission plumes extended beyond the spectrometer’s field of view, and were thus classified as “cutoff” and not quantified. We estimate emission magnitude for these emissions by drawing randomly from the distribution of quantified emissions for well sites and midstream infrastructure, respectively. The number of emission source measurements is not reported for 10 of 11 pipeline emissions in the Kairos Fort Worth survey, out of 72 identified emission sources. We assume these emissions are fully persistent, setting the number of measurements equal to the number of detected emissions at that source.

Step 3: Account for partial detection

For emissions approaching the minimum detection level of an aerial detection system, there may be a fractional probability of detection. If an aerial survey of a population of assets detects an emission of a size that corresponds to a known probability of detection of 1/3, that implies that the survey likely missed two emissions of similar size. Thus, an aerial survey will tend to underestimate emissions in this partial detection range by a predictable amount.

We correct for this effect in the Kairos surveys in the New Mexico Permian basin and the Fort Worth basin, using probability of detection curves based on controlled release testing from ^13,26. See Materials and methods and the SI, Section S1.7 for further detail.

Carbon Mapper has conducted internal controlled release testing to characterize its minimum detection range ³⁷. However, we do not have sufficient single-blind controlled release data to apply a similar correction to Carbon Mapper surveys, many of which were also conducted at varying altitudes, further changing lower detection characteristics. This introduces conservatism into estimates of aerially measured emissions from Carbon Mapper campaigns.

Step 4a: Simulate well site emissions

We then produce a comprehensive well site-level emissions inventory for the surveyed region, as the basis for estimating emissions missed by the aerial survey. We simulate emissions at all surveyed well sites using a basin-scale emissions simulation tool, introduced in ⁵. The bottom-up emissions simulation begins with field measurements of the prevalence and magnitude of emissions at the component level, e.g. valves, flanges, and open-ended lines. It then converts these into probabilistic equipment-level emission factors based on component counts for different types of equipment, e.g. separators, meters, and wellheads.

We update this simulation tool with basin-specific equipment activity data from the EPA’s Greenhouse Gas Reporting Program, e.g. the number of wellheads and pneumatic controllers per site in a given productivity range, as well as production data, to probabilistically estimate emissions at each well site in a given basin. This analysis thus estimates well site-level emission rates for all surveyed active oil and gas well sites in the six basins.

Simulated well site emissions are based on component-level measurements of methane emission frequency and magnitude, combined with counts of the number of each relevant component (e.g. valves, connectors, and open-ended lines) per piece of well site equipment (as listed in the previous paragraph). Eqs. (1) and (2) summarize the underlying mathematics behind this probabilistic emissions estimation method for a given basin, described in detail in the SI, Section S1.4 and ⁵.

Where Q_i is simulated emissions for a given simulated well, i, and Q_basin is methane emissions from all well sites across the oil and gas-producting basin in question. The i index iterates across all wells in the basin, totalling n_wells. The j index iterates across equipment types, with a total of n_equip types. Q_i,j is a randomly-generated equipment-level emission factor for equipment type j at well i, drawing upon empirical measurements of component counts per piece of equipment, the fraction of components emitting at a given time, and component-level emission rates per emission, described further in the Rutherford et al. 2021 and in the SI, Section S1.4 ⁵. is an equipment activity factor (equipment count per well), drawn from EPA GHGRP data for the basin containing the simulated region. Finally, wells are translated into well sites using the spatial clustering algorithm introduced in ⁶. The result is a distribution of well site-level emissions based on the Q_i values.

We identify the number of wells surveyed in a given campaign by filtering the Enverus coordinates of all active wells in the relevant basin by each aerial survey area ³⁸. Enverus does not divide wells into well sites. We convert this count of wells to a count of well sites, assuming the average number of wells per site for the basin, derived from the basin-specific emissions simulation model results, which using the well-to-site clustering algorithm introduced in ⁶. See the SI, Section S1.6 for further detail.

To account for differences in well site productivity between the surveyed area and the basin as a whole, for each campaigns we draw simulated emissions for each surveyed well site from a well site with similar natural gas productivity. This ensures that simulated emissions are representative of the surveyed area, but does not guarantee that the overall emissions estimate from the surveyed area will be representative of the basin as a whole. See the SI, Sections S13, S1.5, S5 for further detail. The result is simulated emission levels for all well sites covered by each aerial survey.

Step 4b: Simulate midstream emissions

We do not have a site-level emissions simulation tool for midstream infrastructure, comparable to the above well site emissions simulation method. We rely on national and state-level Greenhouse Gas Inventory (GHGI) estimates from the United States Environmental Protection Agency (EPA), which includes reported annual values from 2016 through 2020 ^9,25. These estimates are based on similar emissions simulation methods.

EPA’s national inventory includes itemized national emissions from petroleum and natural gas systems. We consider midstream emissions to include EPA’s categories of Gathering and Boosting, Processing, and Transmission and Storage. See Table 2 for national methane emission values by sector by year. We then convert these values into a methane fractional loss rate, dividing by gross onshore US natural gas production drawn from Enverus, excluding production from the federal offshore regions of the Gulf of Mexico and Pacific, converted to methane assuming the same 90% methane fraction assumed in the rest of this work ^6,38. See Table Table 2 for the resulting calculated national methane fractional loss rate by year, from natural gas midstream infrastructure and oil and natural gas production, as well as emissions from natural gas and petroleum sytems as a whole. Because the 2022 GHGI does not include values for 2021, we use 2020 values for campaigns conducted in 2021.

Table 1: National methane emissions from oil and natural gas by sector in the United States from the 2022 EPA Greenhouse Gas Inventory in millions of metric tons of methane per year ²⁹. Includes national onshore methane production from onshore oil and natural gas activity from Enverus ³⁸. National methane fractional loss rate estimates include ±18% error for natural gas system emissions and +32%/-28% error for petroleum systems, derived from reported uncertainty in 2020 ²⁹.

Sector (MMt/yr)	2016	2017	2018	2019	2020
Gathering & Boosting	1.46	1.53	1.55	1.6	1.5
Processing	0.45	0.46	0.48	0.51	0.49
Transmission & Storage	1.53	1.46	1.54	1.58	1.63
Midstream Total	3.44	3.45	3.57	3.68	3.62
Production Total	3.67	3.71	3.66	3.64	3.47
Production + Midstream Total	7.11	7.16	7.23	7.32	7.09
Total methane production	581	600	672	735	730
Methane fractional loss rate midstream + production [%]	1.2 [1.0, 1.5]	1.2 [1.0, 1.4]	1.1 [0.9, 1.3]	1.0 [0.8, 1.2]	1.0 [0.8, 1.2]
NG Total (from exploration to post-meter)	6.61	6.66	6.87	6.89	6.6
Oil Total (from exploration to post-meter)	1.62	1.62	1.54	1.62	1.61
Oil & NG Total	8.23	8.28	8.42	8.5	8.21
Methane fractional loss rate (from exploration to post-meter) [%]	1.3 [1.0, 1.5]	1.2 [1.0, 1.5]	1.1 [0.9, 1.3]	1.0 [0.8, 1.3]	1.0 [0.8, 1.2]

The GHGI also produces state-level emissions inventories ²⁵. These include estimates of total methane production from natural gas systems and petroleum systems, without itemizing midstream and production emissions. For each state, we compute the methane fractional loss rate from natural gas systems by dividing by the state-level GHGI estimate by statewide production from Enverus in the corresponding year, again assuming a 90% methane fraction ^6,38.

We then estimate the methane fractional loss rate from midstream infrastructure in each state by assuming that midstream emissions represent the same fraction of total emissions in each state as they do nationally, 42-44% from 2016-2020 ^6,29. To ensure a conservative estimate, if this value is larger than the national midstream methane fractional loss rate, we use the national rate instead. See Table 2 for estimated natural gas system and midstream fractional loss rates for each state covered in this study. All cases except the CM Permian campaigns cover only one state. In the CM Permian campaigns, we use the midstream methane fractional loss rate from Texas, as this constitutes most assets surveyed. See Table 2 for a mapping between each campaign and the state used to estimate the midstream methane fractional loss rate.

Table 2: Statewide methane emissions from oil and natural gas systems for each campaign from the EPA state-level greenhouse gas inventory. Combined with state-level onshore natural gas production from Enverus. Converted to a midstream loss rate assuming that the national fraction of oil and natural gas system methane emissions holds in each state (42% in 2016, 42% in 2017, 43% in 2018, 44% in 2019). The midstream loss rate is the minimum of this computed rate and the national midstream methane fractional loss rate (0.61% in 2016, 0.60% in 2017, 0.52% in 2019, 0.51% in 2020). The sub-aerial midstream loss rate excludes the 23% of midstream emissions estimated to come from emission sources of 42.7 kg/hr (39.5 scf/min), based on ³⁹ as described in Step 5b. Kairos refers to Kairos Aerospace. CM refers to Carbon Mapper.

Basin/ Campaign	GHGI year	State	Oil & NG CH4 emissions [kt/yr]	CH4 production [kt/yr]	Oil & NG CH4 loss rate [%]	Midstream CH4 loss rate [%]	Sub-aerial midstream loss [%]
Kairos NM Permian	2019	NM	406	33,772	1.25%	0.52%	0.38%
CM Permian/ 2019	2019	TX	2,096	189,440	1.15%	0.50%	0.36%
CM Permian/ 2020	2020	TX	1,948	189,814	1.07%	0.47%	0.34%
CM Permian/ Summer 2021	2020	TX	1,949	189,814	1.07%	0.47%	0.34%
CM Permian/Fall 2021	2020	TX	1,948	189,814	1.07%	0.57%	0.34%
CM San Joaquin/ 2016	2016	CA	349	6,327	5.74%	0.61%	0.45%
CM San Joaquin/ 2017	2017	CA	320	6,114	5.45%	0.60%	0.44%
CM San Joaquin/ Summer 2020	2020	CA	316	5,328	6.16%	0.51%	0.38%
CM San Joaquin/ Fall 2020	2020	CA	316	5,328	6.16%	0.51%	0.38%
CM San Joaquin/ Fall 2021	2020	CA	316	5,328	6.16%	0.51%	0.38%
CM Denver-Julesburg/ Summer 2021	2020	CO	387	41,397	0.97%	0.43%	0.31%
CM Denver-Julesburg /Fall 2021	2020	CO	387	41,397	0.97%	0.43%	0.31%
CM Pennsylvania/ 2021	2020	PA	684	129,151	0.55%	0.24%	0.18%
CM Uinta/2020	2020	UT	101	4,366	2.40%	0.51%	0.38%
Kairos Fort Worth/2021	2020	TX	1,948	189,814	1.07%	0.47%	0.34%

Step 5a: Combine aerially measured and simulated well site inventories

We then combine the generated well site-level inventories of aerially measured and simulated emissions from Steps 2-4a. For each Monte Carlo realization we transition from simulated to measured emissions at the emission size at which measured emissions consistently dominate estimated emissions. This approach avoids double-counting across aerial and simulated emissions inventories. See the SI, Sections S1.8 and S3 for further detail.

Note that this transition point may be larger than the smallest emission detected in the corresponding aerial survey. This is because aerial emissions detection systems generally detect only some fraction of emissions below a certain size. This partial detection range can vary depending on the technology, quality control processes, and environmental conditions. Thus, raw aerial measurements may underestimate total emissions on the low end of the detected range.

For Kairos, we are able to correct the aerially measured emissions distribution for missed emissions in this partial detection range using results from single-blind controlled release testing ²⁶. See the SI, Section S1.7 for further detail. Although Carbon Mapper has conducted some controlled release testing to characterize minimum detection capabilities, we do not have sufficient data to apply similar correction factors, as discussed in the SI, Section S1.1.

Thus, we use measured emissions for all well sites with emissions larger than or equal to the transition point, after accounting for partial detection. We use simulated emissions for all other surveyed well sites.

Step 5b: Combine aerially measured and simulated midstream inventories

Simulated midstream emissions inventories, derived from the EPA GHGI, are not disaggregated into site-level emissions. As a result, we cannot directly apply the above method, removing all simulated emissions above (and all aerially measured emissions below) a transition point emission size.

Instead, we remove a fraction of simulated midstream emissions corresponding to aerially visible emissions from key field measurements that underlie midstream emissions estimates from the GHGI. Midstream emissions estimates in the national GHGI use several data sources, including company-provided submissions to the Greenhouse Gas Reporting Program, as well as peer-reviewed field measurements and simulations ⁹. The GHGI uses Zimmerle et al. 2015 as the basis for estimates of transmission and storage emissions and compressors ^9,39. Zimmerle et al. estimate that 1 in 25 facilities is a “super-emitter”, with emissions of at least 35.9 standard cubic feet per minute (scf/min) of methane, or 42.7 kg/hr, based on the molecular weight of 19.2 grams of methane per standard cubic foot ⁴⁰. The average emission rate of observed emissions in this range is 496 scf/min, or 536 kg/hr per facility ³⁹.

Thus, average super-emitter emissions per facility are 536/25 = 21.4 kg/hr. The minimum emission in this “super-emitter” range, 42.7 kg/hr, is comparable to the transition point between simulated and aerially measured emissions distributions for well sites. For simplicity, our midstream estimates include all aerially measured emissions and remove the estimated share of simulated emissions from sources of 42.7 kg/hr or higher.

Zimmerle et al. estimate average emissions per compressor station at 96.7 kg/hr (847 t/yr), with 76.5 kg/hr (670 t/yr) for transmission stations. If aerially visible emissions contribute an average of 21.4 kg/hr per facility, this represents 22% and 28% of total emissions, respectively. To be conservative, we remove 28% of all midstream emissions, including pipelines, to avoid overlap with the aerially measured distribution. See Table 1 for the simulated midstream methane fractional loss rate excluding aerially detectable emissions.

EPA includes additional low and high estimates of methane emissions from natural gas systems, -18% and +18%, for the year 2020. We apply these uncertainties to midstream emissions estimates when computing the 95% confidence interval for total estimated methane emissions for well sites and midstream assets, using the low value to compute the 2.5^th percentile and the high value to compute the 97.5^th percentile.

We then multiply this methane fractional loss rate by total methane production from all covered well sites to generate an estimated methane emission rate from midstream assets, including only emissions below 42.7 kg/hr, corresponding roughly to emissions below the aerial minimum detection threshold. We compute total methane production from Enverus, including all active oil and natural gas wells covered in the aerial campaign, as described in the SI, Section S1.5.

Step 6: Methane fractional loss estimation

We estimate campaign-specific fractional loss rate, L_c as follows:

Where E_c is total estimated emissions from all well sites and midstream assets covered in the campaign, based on the combined distribution of aerially measured emissions, corrected for partial detection, plus simulated emissions, as described in steps 1-5b. We then compute P_c, total methane production for all covered well sites, using the method described in the SI, Section S1.5.

Note that we then assume this gas has a molar methane fraction of 90% in all basins ⁶, likely a conservatively high estimate more representative of transmission pipeline-ready natural gas than the gross gas production at the wellsite reported by Enverus, discussed further in the SI, Section S7. A lower molar methane fraction would reduce estimated methane production, thus increasing the estimated methane fractional loss rate.

Methods References

33. Thorpe, A. K., Bue, B. D., Thompson, D. R. & Duren, R. M. Methane Plumes Derived from AVIRIS-NG over Point Sources across California, 2016-2017. https://doi.org/10.3334/ORNLDAAC/1727 (2019).

34. Cusworth, D. H. Methane plumes for NASA/JPL/UArizona/ASU Sep-Nov 2019 Permian campaign [Data set]. https://doi.org/10.5281/zenodo.5610307 (2021).

35. Cusworth, D. et al. Methane plumes from airborne surveys 2020-2021 (1.0) [Data set]. https://doi.org/10.5281/zenodo.5606120 (2021).

36. Berman, E. S. F., Wetherley, E. B. & Jones, B. B. Technical White Paper: Methane Detection. Kairos Aerospace (2021) doi:10.17605/OSF.IO/HZG52.

37. Thorpe, A. K. et al. Mapping methane concentrations from a controlled release experiment using the next generation airborne visible/infrared imaging spectrometer (AVIRIS-NG). Remote Sensing of Environment 179, 104–115 (2016).

38. Enverus. Enervus Exploration and Production. https://www.enverus.com/industry/exploration-and-production/ (2022).

39. Zimmerle, D. J. et al. Methane Emissions from the Natural Gas Transmission and Storage System in the United States. Environ. Sci. Technol. 49, 9374–9383 (2015).

40. GPSA. Section 1 General Information. in Engineering Data Book vol. 1 (Gas Processors Supply Association, 2011).

Yes there is potential Competing Interest. The authors declare the following competing financial interest(s): E.S.F.B., P.V.Y., B.B.J., and E.B.W. are employees of Kairos Aerospace. RM.D., D.H.C., and A.A. are employees of the University of Arizona and seconded to the non-profit organization Carbon Mapper. Remaining authors have no competing interests to declare.

Supplementaryinformation.docx

Quantifying oil and natural gas system emissions using one million aerial site measurements

Status:

Journal Publication

Version 1

Abstract

Figures

Main Text

1. Uniting Aerial Measurements With Simulated Emissions

1.1 Methane loss rates vary widely over space and time

2. Aerial Measurements Often Dominate Total Emissions

3. Accurately estimating methane emissions requires measuring almost everything

Declarations

References

Methods

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1