Improving the information content of IASI assimilation for numerical weather prediction

The Infrared Atmosperhic Sounding Interferometer (IASI) provides significant impact to numerical weather prediction systems despite current assimilation schemes using less than 2% of the channels. The current system does not achieve the information content predicted by earlier theoretical studies and results presented here show that the information content could be doubled if the full spectrum were exploited. There is potential to improve the vertical resolution of the humidity analysis and the stratospheric temperature in particular. This thesis explores principal component (PC) compression and radiance reconstruction to compress the spectrum by over 90% whilst retaining almost the full information content. Theoretical calculations are shown that indicate PC scores and reconstructed radiances achieve close to the maximum information content, making them promising approaches for better exploitation of IASI. However, care must be taken because neglected error terms and matrix conditioning are problematic due to the way the information in the compressed observations is coupled in the vertical. New methods for choosing reconstructed radiance channels for assimilation are developed and tested, generating channel selections suitable for implementation in the Met Office operational system. The final section is concerned with the interaction between the observation information and the background error covariance matrix. This matrix can only ever be estimated, which causes the analysis to be suboptimal. If the differences between true and assumed errors are large enough, the analysis may be degraded relative to the background. Guarding against exaggeration of background errors is therefore important, and for water vapour in particular, spurious vertical structures in the stratosphere must be avoided. Increasing the spectral coverage increases the information content and reduces exposure to analysis degradation. This result is encouraging because it means that there is no greater risk to the analysis if more spectral information is provided, paving the way for assimilation of reconstructed radiances.