MSR, a multi-spectrum retrieval technique for spatially-temporally correlated or common Venus surface and atmosphere parameters

Abstract

A common way to regularize mathematical ill-posed retrieval problems in atmospheric remote sensing is the incorporation of single-spectrum Bayesian a priori mean values and standard deviations for the parameters to be retrieved, along with measurement and simulation error information. This decreases the probability to obtain unlikely parameter values. For a reliable evaluation of measurements with sparse spectral information content like Venus' nightside emissions in the infrared as acquired by the VIRTIS-M-IR instrument aboard ESA's Venus Express spacecraft, it can help to consider further a priori knowledge. A new multi-spectrum retrieval technique (MSR) is presented that allows one to incorporate expected correlation lengths and times for the retrieval parameters used to describe several spectra. It is demonstrated by examples that this decreases the probability to retrieve spatial–temporal state vector distributions that are incompatible with these a priori spatial–temporal correlations. Also, a priori correlations between the parameters used to describe a single spectrum and exhibiting similar a priori spatial–temporal behavior, act to rule out unlikely single-spectrum state vectors. Parameters with infinite correlation length or time and identic single-spectrum a priori data are spatially or temporally constant and can be retrieved as parameters that are common to a certain selection of measurements. This is shown to be especially useful to retrieve surface emissivity in the infrared as a parameter that is common to several measurements that repeatedly cover the same target, and to determine deep atmospheric CO2 opacity corrections, which are common to all Venus nightside spectra. Also this way, all considered measurements can be parameterized by a fully consistent set of atmospheric, surface, and instrumental parameters that respects all available a priori data as well as the measurement and simulation error distributions and that does not neglect the context between adjacent measurements. MSR is demonstrated to enhance the retrieval reliability and accuracy and pushes the VIRTIS-M-IR data evaluation to its limits.