British researchers have developed a method. tool to automatically detect columns methane From orbit to Earth using machine learning with hyperspectral data. This could help identify the location of methane “super emitters” so that action can be taken more quickly against these greenhouse gas emissions.
Although climate neutrality goals focus on CO2 emissions, tackling methane emissions is also vital to slowing rising temperatures. And this Methane traps 80 times more heat than CO2although it has a much shorter atmospheric life (about seven to 12 years compared to the centuries that CO2 remains).
Therefore, acting quickly to reduce methane emissions from human sources will have a direct impact on slowing global warming and improving air quality. It is estimated that easily achievable reductions in methane emissions will be achieved. It could prevent almost 0.3°C of warming over the next two decades.
But until now, there have been few methods to easily map methane clouds from aerial imagery, and their processing has been time-consuming. The reason for this is Methane gas is transparent to both the human eye and the spectral ranges used in most sensors satellites. Even when such sensors operate in the correct spectral range to detect methane, data is often obscured by noise and laborious manual methods are required to effectively identify plumes.
The new system is much more efficient
New machine learning tool developed by researchers from Oxford (UK) overcome these problems by detecting methane clouds in hyperspectral satellite data. These detect narrower bands than more common multispectral satellites, making it easier to tune the specific methane signature and filter out noise. But the amount of data they produce is much greater, making it difficult to process without artificial intelligence (AI).
The researchers tested the model using 167,825 hyperspectral mosaics (each representing an area of 1.64 km2) captured by NASA’s AVIRIS weather sensor over the Four Corners region of the United States. The algorithm was then applied to data from other hyperspectrals in orbit. sensors.
Generally, The model has more than 81% accuracy in detecting large methane plumes and is 21.5% more accurate than the most detailed systems previously available.. The method also had a significantly improved false positive detection rate, reducing them by approximately 41.83% compared to the previous system.
To encourage further research into methane detection, the researchers have made both the annotated dataset and the code used for the model available on the project page on GitHub. They are now investigating whether the model can work directly on the satellite.It allows other satellites to make follow-up observations as part of the NIO.space initiative.
Lead researcher, PhD student Vít Růžička (Department of Computer Science, University of Oxford), said: “This built-in process could mean that initially only priority alerts need to be sent to Earth, for example an alert signal with the coordinates of an identified methane source.” This will allow a bunch of satellites to cooperate autonomously“The initial weak detection could serve as a warning signal for other satellites in the constellation to focus their images on the required location.”
Reference work: DOI: 10.1038/s41598-023-44918-6
……………………
Contact address of the environmental department:[email protected]