Bayesian approach to retrieval of vertical ozone profile from radiometry data

A technique for retrieving vertical distributions (profiles) of atmospheric gas constituents from data of passive remote sensing of the atmosphere is proposed. The goal of the technique based on the statistical (Bayesian) approach to solution of inverse problems is construction of probability distribution for a sought quantity throughout the interval of the studied heights. It is assumed that initial data contain measurement noise, and a priori information about properties of the profile is used. It is proposed to approximate the sought profile by a function in the form of an artificial neural network. This approximation allows optimal inclusion of a priori information into retrieval procedure, thus ensuring the most effective regularization of the problem. Efficiency of the proposed technique is demonstrated on an example of retrieval of vertical ozone profile from data of ground-based sounding of the atmosphere in the millimeter wavelength range. Results of profile retrieval from model data and from spectra of radiation temperature of the atmosphere measured in the Apatity (67° N, 33° E) in the winter of 2002-2003 are presented.