The Unmixing of Atmospheric Trace Gases From Hyperspectral Satellite Data

A new approach for the retrieval of the vertical column concentrations of trace gases, from hyperspectral satellite reflectances, is presented. The investigation moves from the general rationale of independent component analysis, but the constraint of perfect independence among sources is replaced by a minimum dependence concept that proves more reasonable for the application at hand. The unmixing of the gas spectra and their concentrations is achieved from linear mixtures obtained from the logarithm of the spectral reflectance. After a proper preprocessing stage aimed at reducing major residual dependences caused by atmospheric scattering, trace-gas retrieval is carried out through a minimization of a statistical cost function, subject to the physical constraint that the resulting spectra must be nonnegative. The experimental analysis relies on the retrieval of sulfur dioxide during volcanic emissions using data from the National Aeronautics and Space Administration Ozone Monitoring Instrument. To validate the procedure, reference reflectance spectra having a known profile of sulfur dioxide are generated with the MODerate resolution atmospheric TRANsmission software, and the retrieved concentration is compared with the theoretical one. Performance in the presence of shot and detector noise has also been analyzed starting from pure simulated spectral reflectances.