In a bid to reduce the damaging impact of ultra emission methane leaks — which are oftentimes traceable to oil rig or natural gas pipeline malfunctions — researchers from the University of Oxford have developed a new open-source tool that marries machine learning and orbital data from satellites to improve the detection of such plumes.

Currently, real-time concentrations of ultra emission methane leaks are difficult to detect both with the naked eye and using most satellites’ multispectral near-infrared wavelength sensors. Yet, reducing such ultra emission methane leaks could help stave off rising temperatures, which are associated with climate change.

Source: University of Oxford

As such, the University of Oxford team developed a new machine learning-powered tool that identifies methane plumes via narrow hyperspectral bands of satellite imaging data. According to the Oxford team, these more specific bands reportedly generate more data.

To train the new model, the Oxford researchers fed it around 167,825 hyperspectral images — obtained from NASA’s Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) satellite — that each measured around 0.66 square miles. The team further trained the model using orbital monitors, like NASA’s hyperspectral EMIT sensor, which is currently aboard the International Space Station.

According to its developers, the new model is about 21.5% more accurate at identifying methane plumes than existing tools.

The tool is detailed in the article, "Semantic segmentation of methane plumes with hyperspectral machine learning models," which appears in the journal Nature Scientific Reports.