Matched-filter trace-gas plume detection

Matched-filter trace-gas plume detection answers the question: is there a localised enhancement of a target gas at this location, and how large is it? The method applies a statistical filter tuned to the spectral absorption signature of a specific gas, extracting the concentration signal from the variable surface background without requiring an explicit atmospheric retrieval model for each pixel.

The column-wise matched filter operates on imaging spectrometer radiance cubes. A target spectrum is derived from the lab-measured gas transmittance (for methane, the SWIR window around 2200-2430 nm; for CO2, around 2020-2070 nm). The background is modelled as a column-wise multivariate Gaussian distribution, fitted per image column to capture the spatial structure of the scene. The filter projects each pixel's spectrum onto the target direction in spectral space, producing a concentration-pathlength enhancement in ppm-m. Integrated emission rates are estimated from the concentration-pathlength map using a cross-sectional flux method.^[1]

The mag1c algorithm (Foote et al. 2020) extends the basic matched filter with L1 sparse regularisation and albedo correction, yielding 60.7% lower RMSE and 2.64x reduction in background noise relative to earlier robust matched filter implementations.^[2] AVIRIS-NG JPL greenhouse gas survey flights have applied this pipeline operationally for methane and CO2 point-source mapping.^[3]

The method requires a VSWIR imaging spectrometer with SWIR coverage to at least 2300 nm for methane, spectral resolution under 10 nm, and SNR above 200:1 in the SWIR for reliable sub-1000 ppm-m retrievals. Detection thresholds are typically 200-500 ppm-m for methane at 3-8 m resolution. Failure modes include low-albedo surfaces (water, dark soils, shadow), surface materials with spectral structure that mimics the target gas absorption, and high aerosol loading.