A team of University of Oxford researchers partnered with Trillium Technologies’ NIO.space to automatically detect methane plumes on Earth from orbit using machine learning with hyperspectral data. The purpose is to reduce methane emissions, which helps slow global warming and improve air quality. While it is essential to reduce carbon dioxide, methane gases from anthropogenic sources are 28 times more potent than carbon dioxide in trapping heat in the earth’s atmosphere. It is the second-largest greenhouse gas (GHG) after carbon dioxide.
Therefore, using this technique could help in extracting information from hyperspectral satellite images. Before this, there were not many methods to map methane plumes from aerial images. If there are any, the technique is very time-consuming. The reason is due to the transparent color of methane gases to the satellite sensor’s spectral ranges and human’s naked eye. With the new machine learning tool these researchers have developed, it will help solve the problems.
These have lower detection bands than more widely used multispectral satellites, which makes it easier to tune in to the exact methane signature and reduce noise. However, processing them without artificial intelligence (AI) becomes challenging because of the amount of data they generate.
Using NASA’s airborne sensor AVIRIS for the model’s training, 167,825 hyperspectral tiles, or 1.64 km2 each, were gathered over the US Four Corners region. Then, the algorithm was applied to the collected data.
This method has been proven to have 81% accuracy compared to the previous one, which was 21.5%. Also, compared with the previous most accurate methodology, this new method’s false positive detection rate for tile categorization was greatly improved, falling by almost 41.83%.
Both the code and the annotated dataset are available on GitHub. The team is investigating satellite on-board processing and how it could be used in the NIO.space project to allow satellite swarms to autonomously collaborate. Lead researcher Vít Růžička imagines that sending priority signals to Earth will increase the effectiveness of methane detection.
