Maximum information composite channels of a high-resolution satellite sounder

Abstract

Recent infrared sounders such as AIRS and IASI have thousands of channels. Because computational efficiency is crucial for practical soundings, it is important to reduce the number of channels that are used. One approach is channel selection, in which a limited number of channels with large information content are selected from among all the channels of a sounder. Here we study another approach. Virtual channels, which are referred as Maximum Information Composite Channels (MICC) hereafter, are constructed by a linear combination of original channels so as to maximize the entropy reduction in atmospheric retrieval. We found that the number of MICCs that had almost the same entropy reduction as the original channels was about a tenth the number of original channels when the MICC method was applied to 185 channels of AIRS. MICCs can be used in the same way as conventional channels in atmospheric retrieval. The channel number reduction leads to a reduction of the matrix size and the computational cost in the retrieval. A numerical experiment on temperature and humidity retrievals showed that the CPU time for the MICC approach was significantly smaller than that of the conventional approach with almost the same retrieval error.