Use of retrievals for data compression and assimilation

The wealth of data acquired by satellite data in the infrared spectral region provides operational meteorological centres with essential observational information so as to constrain initial conditions for prediction. In order to benefit from this abundance of data it is important to put in place appropriate strategies to allow both data providers and data users to use such data in an efficient way. A fundamental difficulty to achieve this objective is that the problem of determining the model state vector that is consistent with the available measurements is in general a nonlinear inverse problem. In this particular application, nonlinearity of the observation operator is usually due to phase transitions of atmospheric water. The solution of this kind of problems may require sophisticated inverse methods that are not suitable for operational implementation. In this case, a possible way forward is to apply a preliminary inverse method to find a solution (or retrieval) around which the inverse problem becomes linear. In this case it is possible to reduce the data while preserving most of its information as well as simplify significantly its assimilation in NWP models. In this talk, possible strategies to achieve this result are discussed, with particular emphasis on both the advantages as well as potential difficulties of this method as compared with alternative techniques.