TRACKING THE INVISIBLE ENEMY: METHANE DETECTION USING MACHINE LEARNING METHODS

Detecting methane in airborne and satellite hyperspectral images (HSIs) can play a crucial role in environmental monitoring, as timely and spatially-scalable actions to reduce emissions and address unexpected super-emitters are of critical importance. We tackle this challenge and propose a machine learning process—coupling unsupervised clustering for determining superpixels with supervised models for their later classification—for the detection of methane, with the ultimate goal of making it flexible and deployable on board a satellite. Such a solution not only potentially offers global scalability, but can act as an intelligent data prioritization step, as only HSI containing methane can be transmitted for further analysis. Our experimental study performed over the STARCOP dataset demonstrated that the suggested pipeline can precisely detect methane in a limited number of spectral bands.