Description of the upgrade

1. two modifications of the forecast error penalty function (Jb). One allows the use of a refined balance constraint, with most impact in the tropics (better fit of the analysis to the observations). The result is also slightly positive in terms of forecast scores. The other modification is in the formulation of the representativeness error growth used for the assimilation of asynoptic data. The errors are now supposed to saturate at levels that are fully varying with space and this was shown to have a positive impact both on the analysis increments and on the data quality control;
2. several modifications in the way the TOVS data are used (quality control, modification of the tropical bias correction in line with the Jb re-formulation);
3. a restriction of the dynamic initialisation to the small scales.