Super Channels for AIRS Retrievals (Invited Presentation)

Hyper spectral instruments such as AIRS and IASI have the order of 100 times more channels than those of current instruments. The larger number of channels poses a computational challenge for using these measurements in a numerical model because radiances must be calculated for all the channels. The large number of measurements is generated only because the high spectral resolution that produces them is needed to produce atmospheric weighting functions with the required vertical resolution, not because all the channels are required. Many of the channels sense similar regions of the atmosphere and have the same response to atmospheric changes. In a noise free environment, once the first channel is obtained, the others would provide no additional information. When noise is present, use of all the channels reduces the noise. This suggests averaging similar channels as a way to reduce the number of calculations while preserving the information content. A common way that has been used to compact the date is the eigenvector approach. However, this does not solve the computational problem for the numerical models because the available rapid transmittance approaches cannot be applied to eigenvectors.

The super channel approach consists of averaging channels that have a similar spectral response. Forming the super channels is only one of three requirements that must be satisfied in order to reduce the number of calculations. The second requirement is to have a rapid method for calculating transmittances for the super channels. The third requirement is to be able to generate a Planck function equation for the super channels. Once all three requirements are met, radiances can be calculated. The calculation effort for a single super channel is then the same as is required for calculating a single AIRS channel. It takes about 300 super channels to represent the information content of all the 2378 AIRS channels. About 100 hundred of these are unique channels that can't be combined with anything else and are not averaged. Many of these are combinations of gases such as water vapor and dry channels that have little application to producing retrievals and may not be required, making the process even more efficient. The errors of the various steps and the information content that is preserved are shown.