Satellite microwave observations are still not fully used over land within assimilation systems in spite of their large information content. So far, only observations that receive the least contribution from the surface are assimilated. New developpements have been carried out at CNRM/GAME to extend the use of microwave data to the land surface. Three methodologies have been developed for AMSU-A, AMSU-B and SSM/I (Karbou et al. 2006), and have been implemented in the ARPEGE 4D-Var assimilation. The first method (1) uses averaged emissivity atlases derived from microwave surface channels averaged over 2 weeks prior to the assimilation period. The second method (2) uses emissivity estimates dynamically derived, within the assimilation system, for each atmospheric situation and for a selction of microwave channels. For AMSU-A, emissivity is calculated at 31GHz and then given the remaining channels. For AMSU-B, emissivity is calculated at 89GHz and then allocated to the other AMSU-B channels. For SSM/I, the emissivity is calculated at 19V and 19H: channels with a horizontal (vertical) polarization use 19H(V) emissivities.The third method (3) is based on the first one with dynamical skin temperature estimation using one surface channel. Of course, channels that are used to calculate the emissivity or the skin temperature are discarded from any other computation or diagnostic to ensure that the same information is not used twice during the assimilation.

The new methods have been tried in the French 4D-Var system and their results have been compared with those obtained using the operational system. All land schemes have been evaluated by examining the performances of the 4D-Var observation operator prior to the assimilation. The performances of the observation operator have been examined in the sense of two diagnostics: (a) observation departure from first guess and (b) number of observations that could be used. The results of the comparison show a significant improvement in the fg-departure statistics (mean/std) when the surface is updated with the new surface schemes. The study indicates that additional AMSU-A, AMSU-B and SSM/I channels could indeed be assimilated, provided an adequate land configuration was chosen. AMSU-A channels 2 to 4 and 15 as well as AMSU-B channel 2 present satisfactory statistics that could allow their use in the 4D-var system. However, if an experiment allows the use of a greater number of observations, it is does not necessarily improve the forecast skills. Therefore, additional analysis should be done in order to give general conclusions.

4D-VAR experiments at ECMWF have been conducted to improve the assimilation of temperature sounding observations from SSMI/S over land. Land surface emissivity and/or skin temperature have been determined using brightness temperatures from a selection of window channels, and the estimates used to assimilate data for higher frequency channels. In addition to SSMI/S experiments, other assimilation experiments have also been performed to assimilate AMSU humidity and temperature sounding channels over land within the IFS system. For all assimilation experiments, the RTTOV simulations are improved when updated land emissivities are used and the assimilation system benefits from a larger number of observations compared to the control. The impact on forecast scores is positive for the Southern Hemisphere and neutral for the Northern Hemisphere when assimilating SSMI/S temperature sounding observations over land and over sea. A similar impact was observed for an experiment that updates the emissivity calculations for AMSU-A and AMSU-B over land (Karbou et al. 2007).