Autoregressive smoothing of GOMOS transmittances

The nominal processing chain of the GOMOS star occultation data follows a two-stage inversion scheme: firstly, the transmittances are spectrally inverted to derive the slant path optical thicknesses of different absorbing species; secondly, the latter quantities are vertically inverted for retrieving number density profiles. Although this algorithm is cheaper than reverting the order of inversions, it neglects the use of the high correlation between successive tangent altitudes. As GOMOS may suffer from residual scintillation effects from oblique occultations, we propose to filter the transmittances along the vertical direction by using the optical segment kernel coupled to a standard regularization operator. The tuning parameter of the filtering operator is optimized by minimizing the Durbin-Watson statistics which ensures a maximal decorrelation of residuals. The results will be compared with the generic data processing model and also with a global inversion scheme.