Retrieval of temperature vertical profile from radiance spectra by inversion of the radiative transfer equation

Inversion of the radiative transfer equation to retrieve the vertical profile of temperature from high resolution radiance spectra is an important problem in remote sensing of atmosphere. Because of its non linearity and ill conditioning, regularization techniques have been resorted in order to reduce the error of the retrieval. In this paper Generalized Singular Value Decomposition (GSVD) and Truncated Generalized Singular Value Decomposition (TGSVD) have been used to solve the linear model; the optimal regularization parameter for the proper amount of smoothing have been chosen by the L-curve criterion. A significant test problem has been worked out with reference to the Infrared Atmospheric Sounding Interferometer (IASI). The effectiveness of the methods to reduce variance and bias in the output profile has been addressed. We show that GSVD plus L-curve criterion or TGSVD plus L-curve are really effective in reducing error, variance and bias of the retrieved profile.