The effect of variation of principal components on retrieval of atmospheric profiles from hyperspectral infrared sounder data

The impact of principal component (PC) temporal and regional dependency on the performance of atmospheric retrieval from hyperspectral infrared sounder observation was investigated. The investigation was conducted for retrieval of temperature and water content vertical distribution from AIRS radiance data using the one-dimensional variation method (1D-Var). Simulation results indicated that by using a suitable number of the leading PCs, we can achieve the same level of retrieval accuracy as all the original 185 channels independently of season and region with much fewer channels. An experiment using real observational data showed that no significant difference in retrieval accuracy due to seasonal and regional variation of PC could be seen, which supported simulation study, but the retrieval using more PCs does not always achieve better accuracy, which was different from the results of the simulation study.