A principal component based version of the RTTOV fast radiative transfer model

A new version of the RTTOV fast radiative transfer model has been developed that exploits principal component analysis. The model is based on the computation of a database of line-by-line spectra for a large training set of diverse atmospheric situations. The principal component scores obtained from the eigenvectors of the covariance matrix of the simulated radiances are used as input data in a linear regression scheme where they are expressed as a linear combination of profile-dependent predictors. The predictors consist of a selected number of polychromatic radiances computed using the standard RTTOV fast transmittance model. The linear regression scheme can then be used to simulate principal component scores and consequently reconstruct radiances for any input atmospheric profile. The principal component based model compares favourably to the conventional RTTOV model both in terms of speed and accuracy: the dimensionality reduction inherent in the use of principal component analysis makes the principal component based RTTOV much more computationally efficient, and because of the highly linear relationship between the principal component scores and the independent variables used in the regression scheme, the principal component based RTTOV can reproduce the underlying line-by-line radiances to a much higher degree of accuracy. The availability of a principal component based fast radiative transfer will enable ECMWF to exploit the noise reduction capability of principal component analysis and investigate the direct assimilation of IASI principal component scores in the spectral regions affected by high instrument noise. Copyright (C) 2010 Royal Meteorological Society