Improved methane flux estimation from hyper-spectral imagery via log-domain matched filtering and background homogenization

New satellite hyper-spectral sensors, such as PRISMA of the Italian Space Agency, observe large portions of the Earth’s surface at visible and at shortwave infrared wavelengths with high spectral and spatial resolutions, enabling the investigation of individual molecular species and the localization of emission sources. The ‘Matched Filter’ (MF) methodology, widely exploited in the methane source identification and in the estimation of enhanced concentrations, is discussed in its theoretical foundations, revised and extended within an integrated processing framework. We apply an estimator (termed MF-EVO) operating in the logarithmic radiance-ratio domain, i.e. optical depth space, which allows to overcome the limitations imposed by the linearization assumption of the classical MF and improves robustness across a wide range of methane concentration enhancements. Results from MF-EVO are compared to the traditional algorithm for a set of synthetic PRISMA observations accounting for both homogeneous and heterogeneous background conditions. The MF-EVO algorithm demonstrates superior performance over the MF-Classic method in identifying methane sources across all idealized conditions. Specifically, the estimated identification limit for ΔXCH4 is approximately 0.05 ppm for MF-EVO, significantly lower than the 0.09 ppm limit for MF-Classic. Furthermore, the MF-EVO consistently outperforms the classic MF in the accurate estimation of concentration enhancements across both small and medium-to-large methane concentration scenarios. Under idealized conditions, MF-EVO achieves an error margin within 5%, which is a substantial improvement compared to the 10%–50% error range observed with the MF-Classic method. To address the challenges posed by real-world scenes, the revised MF formulation is embedded in a processing chain that includes false-positive pixel elimination and scene homogenization through image partitioning into spectrally homogeneous clusters. These steps significantly reduce background-induced artifacts and stabilize methane enhancement retrievals, enabling more reliable plume identification and flux estimation. In the application to the Mumbai metropolitan landfills, the full processing chain reduces the estimated methane fluxes by approximately 40%–55% with respect to the classical MF applied to the full scene, highlighting the impact of background homogenization and false-positive suppression on flux estimation in heterogeneous environments.