Large-Scale Controlled Methane Releases for Satellite-Based Detection and Emission Quantification of Point-Sources: A Master's Thesis

Space-based remote sensing offers a promising tool for consistent monitoring and detection of large methane sources worldwide. Numerous instruments and algorithmic approaches for methane point-source detection and quantification are in place or under development; however, very few publicly available ground-truth datasets exist to validate performance of these systems. Moreover, several prominent satellite platforms launched in the past three years have not yet been tested under single-blind protocol to compare their measurement capabilities against well-characterized and controlled sources. We introduce a novel dataset comprising 289 controlled methane releases conducted under both cooperative (unblinded) and single-blind study design. The releases occurred at two locations in the Western United States between August and March 2025 and were enabled by a fully automated and remote-controlled release system of high-pressure gas. This setup is designed to mimic short-duration, high-emission events from emitting infrastructure, such as an unlit flare from fossil fuel production or an emergency pressure release event. The releases coincide with overpasses from 12 satellites and one airplane instrument. These instruments include government, non-profit, and private missions that were purpose-built or repurposed for detecting methane point sources. This study constitutes the largest collection of large-scale, controlled methane releases to date to the authors’ knowledge. The dataset includes high-resolution time series of gas flow rates (1s) and meteorological measurements during each release, including multi-height wind speed (5s), wind direction, temperature, humidity, soil moisture, solar irradiance, cloud cover, and sky images (~30s). We anticipate that these data will support the development, training, and benchmarking of methane detection and quantification algorithms. In addition to presenting the dataset, we evaluate the detection and quantification performance of satellite analysis teams under both unblinded (5 teams) and single-blind (12 teams) phases of the study. The multi-month campaign included 132 single-blind controlled releases and generated 571 total team responses. Of the non-filtered responses, 73% correctly identified the presence or absence of emissions, indicating good overall detection performance. Quantification results showed relatively unbiased global performance, though precision and accuracy results varied significantly. Cloud filtering thresholds differed notably among teams using the same instrument, further underscoring the role of analyst-specific choices. Near-coincident satellite observations emerged as especially valuable for intercomparison—highlighting a promising area for future research.