Detection of methane plume emissions with satellite imagery

This thesis addresses the detection of methane plume emissions using satellite imagery. The detection of methane (CH4) leaks from anthropogenic activities is a cost-effective and global approach, which is able to rapidly reduce greenhouse gas (GHG) emissions. In this thesis we study matched filter techniques. We formalize the concept of matched filter adjustment factor. This leads to the introduction of two new detectors: the Model Adjusted Matched Filter (MAMF) and the Model Adjusted Generalized Likelihood Ratio Test (MA-GLRT) which are based on the computation of an appropriate adjustment factor. We then exploit the complementarity between the spectral information and the spatial information with the fusion between our detectors and a deep learning model. The second part of our study deals with the detection of plumes with the use of time series. We start by using time series to complete an atmospheric absorption model inversion technique. In particular, we show that this method allows for automatic methane plume detection while also enabling the detection of plumes that are missed by the current state-of-the-art method. We then use time series to perform background subtraction. This background subtraction is completed by pattern recognition techniques: a first one based on the detection of local extrema in a pixel's spectrum and a second one based on the matched filter.