THE USE OF METEOSAT WINDS WITH QUALITY INDICATORS WITHIN THE MÉTÉO-FRANCE GLOBAL NWP MODEL

The use, in a global uniform version of the ARPEGE model, of the Atmospheric Motion Vectors (AMV) produced by EUMETSAT, from Meteosat 5 and 7 platforms, in BUFR form, is investigated. These data are produced every 90 minutes with a Quality Indicator, coming from an Automatic Quality Control (AQC) performed by EUMETSAT. The study of the quality against the background (short range forecast) indicates a potential to use the QI, as a selection criteria within the screening of the data in the analysis. Several experiments are investigated, and compared with the current operational NWP system, which uses only the AMV in SATOB format, a subset of higher quality than the BUFR AMV, but disseminated every 6 hours only. A first experiment, where all observations are used, leaving the choice of active data to the screening procedure during assimilation (operational asymmetric first guess check) appears not to be good enough. Three other experiments (one with some thresholds of QI relaxed at high levels and at mid-levels over the sea in the mid-latitudes, another with the AQC only as selection criterion, and the last one, with some thresholds more restricted than in the previous experiment) give similar results, against the operational NWP system: • a reduction of root-mean-square (rms) increments at high levels over West Africa and in the Southern Hemisphere on the Meteosat coverage, and at low levels in the Southern Atlantic. • a global neutral impact in forecasts, but slightly positive over Europe and Australia. The third experiment (with restricted thresholds), giving a better fit of BUFR AMV to first guess and analysis, appears to be the most robust. Finally, a last experiment, as the third but with a blacklist of data in extra-tropics mid-levels, very biased in zonal component, gives the best result, whilst increasing the number of AMV from Meteosat 5 and 7 for the assimilation by a factor 5, against the operational use of SATOB data. 1