UNLIMITED Developing Methane Emissions Inventories by Fusing Airborne, Satellite, and Modeled Assessments: Comprehensive Surveys of the Anadarko and Haynesville Basins Elena SF Berman (PI), Erin Wetherley, Petr Yakovlev, Evan Sherwin and Adam Brandt March 20 2026 U.S. Department of Energy National Energy Technology Laboratory DE-FE0032310 Project Period of Performance (Start Date: 10/1/2023, End Date: 09/30/2025) Recipient: Kairos Aerospace ​ ​ ​ ​ ​ ​ ​ ​ Insight M ​ ​ ​ ​ ​ ​ ​ MiQ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ Stanford University ​ Lawrence Berkeley National Laboratory ACKNOWLEDGMENT OF FEDERAL SUPPORT AND DISCLAIMER: This material is based upon work supported by the Department of Energy under Award Number DE-FE0032310. This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof As noted in our previous reports, please note that Kairos Aerospace, Inc. officially changed its name to Insight M, Inc. although we continue to do business as Kairos Aerospace under a DBA BODY MATTER 1. Introduction Vented and fugitive natural gas from the oil and gas industry represent a significant loss of valuable energy commodities. Capturing this lost gas is vital to supporting American economic competitiveness, maximizing domestic energy production, and expanding liquified natural gas (LNG) exports. Preliminary analyses suggest that millions of dollars worth of natural gas are lost annually due to correctable equipment and operational failures. By generating actionable insights into the root causes of these losses, oil and gas operators can leverage these data to decrease the proportion of natural gas vented into the atmosphere and bring more U.S. gas to market. Effective detection strategies must weigh tradeoffs between sensitivity, scalability, and frequency, especially because a small percentage of anomalous sources (0.1-20%) contribute a disproportionately high percentage (50-90%) of total uncaptured gas (e.g. Allen et al., 2013; Brandt et al., 2016; Zavala-Araiza et al., 2015; Omara et al., 2018). Historically, generating basin-specific loss inventories has been challenged by disagreements between bottom-up approaches (which rely on component-level equipment factors) and top-down approaches (which use site-level measurements from planes or satellites). Top-down approaches using newer technologies typically find substantially greater gas loss rates than bottom-up estimates. This project aimed to bridge that gap in two critical U.S. regions: the Anadarko (Woodford Shale) and Haynesville basins, both of which are key suppliers to LNG export terminals and vital to local and national economies. 2. Methods, Assumptions, and Procedures To construct comprehensive methane loss inventories, we utilized a coordinated, two-pronged methodology: 2.1 Top-Down Aerial and Satellite Surveys Top-down data collection relied on Insight M's LeakSurveyor™ light-aircraft mounted gas imaging system, paired with GHGSat's high-resolution hyperspectral imaging satellites. The total achieved non-unique (i.e. counting overlapping areas) coverage area of 26,664 sq mi significantly exceeds the targeted non-unique coverage of 9,572 sq mi. ●​ 2023 Basin-Wide Aerial Surveys (Figure 1): We conducted comprehensive aerial surveys covering >70% of total hydrocarbon production in the target basins. DOE funded surveys covered 1,232 sq mi in the Anadarko Basin (Woodford Shale), and 4,077 sq mi in the Haynesville Basin. Insight M donated 9,821 sq mi in the Anadarko Basin (Woodford Shale), and 6,816 sq mi in the Haynesville Basin as cost share. These values represent total sq miles covered across all surveys (i.e. non-unique coverage), and include overlapping regions. Total non-overlapping (i.e. unique) survey areas for 2023 are 5,209 sq mi in the Anadarko Basin (Woodford Shale) and 4,398 sq mi in the Haynesville Basin. We ensured that the sample was representative of oil and gas production in both regions by selecting sites that had a similar distribution of site-level oil and gas production to the basin as a whole. Surveys represented 84% of all wells in the 6 Haynesville Basin, covering roughly 75% of total gas production in both regions. This large sample allowed us to robustly estimate overall gas loss dynamics despite high intermittency. ●​ 2023 Targeted Satellite Surveys: In order to constrain the duration of the largest emission sources, we targeted aerially detected emissions for follow up with GHGSat satellite surveys. ●​ 2024 Basin-Wide Aerial Surveys: Insight M continued donating survey data from commercial operations as part of cost sharing efforts for this project. This included 6,895 non-unique sq mi of coverage in the Anadarko Basin (Woodford Shale) and 5,165 non-unique sq mi in the Haynesville Basin. ●​ Year 2 (2024) Coordinated Temporal Campaign (Figure 2): In Year 2, the protocol was expanded to better capture emission duration. We implemented full satellite surveys over highly representative 350-square-mile sub-regions in both basins. DoE funded aerial surveys totaled 3,066 total non-unique sq mi in the Anadarko Basin (Woodford Shale), and 1,931 non-unique sq mi in the Haynesville. Simultaneous to satellite overpasses, 4 aerial surveys were conducted across a 3-month study period at approximately 4-week intervals. ** All units are Achieved Survey Achieved Survey Total Achieved Target Survey square miles Coverage Area Coverage Area Coverage Coverage Area (DoE- Funded) (Donated Data) Area (2023 & 2024) (2023 & 2024) Haynesville 6,008 11,981 17,989 3,443 Basin Woodford Shale 4,063 16,716 20,779 6,129 (Anadarko) TOTAL 10,071 28,697 38,768 9,572 Table 1 - Completed and targeted Insight M survey areas for 2023 and 2024. Coverage areas represent total sq miles covered across all surveys, and include overlapping regions. 7 Figure 1 - Regions surveyed by Insight M in 2023. Figure 2 - Regions surveyed by Insight M in 2024. 2.2 Emissions Modeling and Midstream Asset Mapping We expand upon the methodology developed by Sherwin et al., (2024) to produce complete methane emissions inventories for the Anadarko and Haynesville basins. This relies on the bottom-up modeling approach developed by Rutherford et al. (2021), which we translated into Python and renamed as the Bottom-up Analyzer for Source Emissions (BASE) model. Outputs from BASE are integrated into the new Regional Oil and Gas Aerial Methane Synthesis Model (ROAMS) model, which has been translated into Python. Both the BASE and ROAMS models have been made publicly available on GitHub. Historically, midstream sites (gathering, boosting, and processing) are poorly mapped. We employed AI computer vision algorithms developed by Stanford University to identify compressor stations and tank batteries from optical imagery collected during our aerial surveys (Figure 3). The AI computer vision results were validated by experts in oil and gas imagery 8 analysis from Kairos Aerospace. These asset inventories, and emission factors informed by data shared by MarkWest Energy Partners, L.P. (MPLX) were then incorporated into the basinwide emissions inventory. Figure 3. Map of all compressor station locations identified by the machine learning algorithm prior to validation for the Haynesville (far left) and Woodford (center) study regions. Note that we expanded the area of model application to a larger region around the study areas. An example detection is shown on the right. Map sources: Open Street Map, Google Maps. 3. Results and Discussion 3.1 Top-Down Detections The 2023 basin-specific aerial surveys yielded 1,112 emission detections across 709 sources, with a total emission rate of 324,111 kg/hr. In the Haynesville basin, aerial surveys detected 848 emissions from 505 sources, with rates ranging from 3 kg/hr to 7,418 kg/hr (median: 136 kg/hr). In the Anadarko basin, aerial surveys detected 264 emissions from 204 sources, with rates ranging from 19 kg/hr to 1,405 kg/hr (median: 96 kg/hr). Of the 52 images collected by GHGSat for follow up on aerially detected emissions, none were able to detect methane plumes from the targeted locations. However, these images included 28 emission detections from other sources, ranging from 213 to 2,153 kg/hr (Figure 4). Figure 4 - Emissions detected by GHGSat in 2023 (Follow up survey group) and 2024 (High frequency survey group). 9 To better characterize the duration and temporal persistence of large gas losses, the Year 2 protocol implemented a high-frequency survey campaign over representative sub-regions. During this campaign, repeated aerial surveys detected 134 individual uncaptured gas emissions across the two basins (Figure 5). In the Haynesville study area, 74 emissions were detected ranging from 5.6 kg/hr to 297 kg/hr (median: 73 kg/hr). In the Woodford study area, 60 emissions were detected ranging from 12 kg/hr to 2,990 kg/hr (median: 85 kg/hr). Simultaneously, the coordinated GHGSat satellite taskings collected 93 large-tile (FOV) observations, capturing 11 distinct large-volume gas loss observations across the target areas, observing emission rates ranging from 163 kg/hr to 4,149 kg/hr (Figure 4). By layering these high-frequency airborne and satellite passes, the project successfully tracked the temporal persistence of equipment failures over the 3-month period. For example, the campaign recorded one persistent upstream source that was actively emitting ~110 kg/hr across 8 separate observations, and another emitting ~75 kg/hr across 5 separate observations. Aerial survey data donated as part of cost share in 2024 yielded an additional 228 emission detections from 201 sources in the Anadarko basin, and 601 emission detections from 416 sources in the Haynesville basin. The combined 2024 aerial survey dataset includes coverage of ~62% of gas production in the Anadarko basin, and ~40% of gas production in the Haynesville basin. Figure 5 - Emission characteristics of aerially detected emissions in the 2024 high frequency survey campaign. 3.2 Basin-specific Emissions Inventories Basin-specific emissions inventories combine bottom-up emission modeling from the BASE model with top-down emission detections via the ROAMS model. We generated bottom-up emissions inventories using the BASE model and emission factors developed for this project (Figure 6). Results from bottom-up modeling indicate that the contribution of emissions from tanks in the two basins is quite similar, while the contribution from pneumatic controllers is 10 higher in the Haynesville. Top-down emissions inventories indicate that the Haynesville and Anadarko basins have contrasting emissions dynamics: the Haynesville loses more gas, but these losses represent a smaller percentage of production volume than the Woodford (Figure 7). In both basins, measured top-down emissions (as a % of total volume of methane produced by the surveyed assets) from upstream and midstream sources are approximately equal, with slightly greater losses from upstream sources. However, differences include the proportion of emissions from pipelines, (representing 22% of total emissions in the Woodford while only 5% in the Haynesville) and the proportion of emissions from compressor stations (representing 16% in the Woodford and 25% in the Haynesville) (Figure 8). Joint top-down and bottom-up inventories indicate that while aerially detected emissions are sourced from 0.38-1% of wells, they represent 50%-60% of the total emitted methane (Figure 9). Figure 6 - Equipment-specific inventories for project regions based on bottom-up emissions factors developed for this project. Note that top-down emission detections are not included in this tabulated data. Figure 7 - Basin-specific emission aerial emissions inventories for the Anadarko (Woodford Shale) and Haynesville basins. 11 Figure 8 - Contribution of individual aerially detected emission source types for the Anadarko (Woodford Shale) and Haynesville basins. Figure 9 - Joint top-down (aerial) and bottom-up (simulated) emission inventories for the Anadarko (Woodford) and Haynesville basins. 4. Dissemination Project findings have been disseminated to the scientific and industrial communities through presentations at the 2024 NETL meeting, the 2024 MIT A+B Conference (and the associated conference proceedings, Kamdar et al., 2024), the 2024 and 2025 American Geophysical Union (AGU) Annual Meetings, and a manuscript published in Nature Communications (Burdeau et al. 2025). 5. Project Termination and Incomplete Objectives Despite the early termination of the grant by the DOE at the conclusion of Year 2, the project team was able to complete all the milestones for years 1 and 2 for the project and to make 12 progress on the majority of the milestones for year 3 of the project, see Table 2. Due to the early termination the project was unable to complete its final dissemination phases and the objective of establishing a full three-year multi-season time series. Table 2 - Status of tasks and milestones planned in this project. 6. Conclusion The project successfully created reconciled, basin-specific inventories for both the Anadarko and Haynesville basins. The deployment of spatially comprehensive airborne sampling proved that large sample sizes and careful spatial stratifications are critical for producing unbiased estimates of heavy-tailed equipment failures. Furthermore, the project successfully mapped previously under-documented midstream infrastructure and integrated these into the methane emissions inventories. By merging these datasets via the newly developed, open-source ROAMS model, the project successfully bridged the historical gap between top-down observations and bottom-up simulations and provided a valuable new open source tool to the entire community. Ultimately, this framework provides oil and gas operators with highly scalable, 13 economically valuable insights to inform operational improvements and bring more natural gas to export and domestic markets. 7. Recommendations Based on the methodologies proven during this performance period, we recommend the following to support American energy competitiveness and reduce natural gas loss: 1.​ Expanded Surveys to Boost Gas Recovery: The survey methodology developed under this grant is highly scalable. We recommend extending this tiered top-down and bottom-up framework to the rest of the United States to improve oil and gas operations, thereby bringing more natural gas to market and improving worker safety. 2.​ Utilization of ROAMS: Operators should utilize the newly open-sourced ROAMS model to integrate top-down measurement data with bottom-up equipment factors, ensuring that small-scale, persistent gas losses are accurately tracked and mitigated. 3.​ Midstream Data Sharing: The integration of proprietary equipment, operational, and gas composition data from operators (such as the MPLX partnership in this project) proved vital to closing the midstream data gap. We recommend continuous industry data sharing agreements to refine loss estimations for compressor stations and gathering pipelines, ultimately improving system-wide efficiency. 8. References Allen, D.T., et al. (2013) Measurements of methane emissions at natural gas production sites in the United States. PNAS 110, 17768–17773. https://doi.org/10.1073/pnas.1304880110 Brandt, A.R., Heath, G.A., Cooley, D., (2016) Methane Leaks from Natural Gas Systems Follow Extreme Distributions. Environ. Sci. Technol. 50, 12512–12520. https://doi.org/10.1021/acs.est.6b04303 Burdeau, P.M., et al. (2025) High-resolution national mapping of natural gas composition substantially updates methane leakage impacts. Nat. Commun. 16, 11297 Kamdar, H., et al. (2024). Creating Accurate Methane Emission Inventories through Data-Driven Airborne Survey Strategies. Applied Energy Symposium: MIT A+B. Omara, M., et al. (2018) Methane Emissions from Natural Gas Production Sites in the United States: Data Synthesis and National Estimate. Environmental Science & Technology 52. https://doi.org/10.1021/acs.est.8b03535 Rutherford, J.S., Sherwin, E.D., et al. (2021). Closing the methane gap in US oil and natural gas production emissions inventories. Nature Communications, 12, 4715. Sherwin, E.D., et al. (2024). US oil and gas system emissions from nearly one million aerial site measurements. Nature, 627, 328–334. 14 Zavala-Araiza, D., et al. (2015). Reconciling divergent estimates of oil and gas methane emissions. PNAS 112, 15597–15602. https://doi.org/10.1073/pnas.1522126112 Zimmerle, D., et al. (2022). Modeling air emissions from complex facilities at detailed temporal and spatial resolution: The Methane Emission Estimation Tool (MEET). Science of The Total Environment, 824, 153653. 7. Appendices Appendix A: Open-Source Code Repositories Developed as Part of this Grant Open-Source Code Repository for ROAMS Model: https://github.com/LBNL-ETA/ROAMS Open-Source Code Repository for the BASE Model: https://github.com/pburdeau/BASE 15