Hyperspectral data noise characterization using principle component analysis: application to the atmospheric infrared sounder

Exploiting the inherent redundancy in hyperspectral observations, Principle Component Analysis (PCA) is a simple yet very powerful tool not only for noise filtering and lossy compression, but also for the characterization of sensor noise and other variable artifacts using Earth scene data. Our approach for dependent set PCA of radiance spectra from the Atmospheric Infrared Sounder (AIRS) on NASA Aqua is presented. Aspects of the analyses include 1) estimation of NEDT and comparisons to values derived from on-board blackbodies, 2) estimation of the signal dependence of NEDN, 3) estimation of the spectrally correlated component of NEDT, 4) investigation of non-Gaussian noise behavior, and 5) inspection of individual PCs. The results are generally consistent with results obtained prelaunch and on-orbit using blackbody and space view data. Specific findings include: 1) PCA estimates of AIRS spectrally random and spectrally correlated NEDN compare well with estimates computed from blackbody and space views, 2) the signal dependence of AIRS NEDN is accurately parameterized in terms of scene radiance, 3) examination of the reconstruction error allows non-Gaussian phenomenon such as popping to be characterized, and 4) inspection of the PCs and filtered spectra is a powerful technique for diagnosing variable artifacts in hyperspectral data.