Abstract

With the advance of hyperspectral sounders such as NAST-I, AIRS (2002), CrIS (2006), IASI (2006), GIFTS (2005/6) and ABS/HES (~2010), hyperspectral noise filtering has become a new challenge for improving atmospheric and surface retrieval. A wavelet-based noise filtering scheme is applied to hyperspectral observations. The optimal threshold for each wavelet subband minimizes the mean square error of the result as compared to the unknown, exact data and is determined by the method of generalized cross validation (GCV). The wavelet transforms used here are fast and are done in-place without additional memory needed. Also, the noise filtering scheme has low complexity for hardware implementation and does not rely on an estimate of noise equivalent radiance (NER). Colored or white noise can be filtered with this scheme. Furthermore, it can be embedded in a wavelet-based hyperspectral lossy data compression scheme. These features have made it an ideal candidate for both onboard and ground hyperspectral noise filtering.

© 2003 Optical Society of America