All observations which are assimilated by a numerical weather prediction are controlled on the basis of statistics relating to the difference between the observation and the model. Thus, each observations available and useful for ARPEGE model is compared with a six hours range forecast (first-guess) interpolated at the point of observation and for the same validity date. Observations whose statistics deviate notably from the average behavior of the other observations compared to the first-guess are regarded as suspect.

METEO-FRANCE produces monitoring statistics on the quality of observations received at Toulouse and which are assimilated by ARPEGE model.
The following site provides information on various types of observation and are of several types :

data coverage maps : distribution of available and used observations on the globe

DFS : "Degree of Freedom Signal" which quantify the power of different observation types to modify assimilation system
FSOi : "Forecast Sensitivity to Observations-based impact" which quantify the power of different observation types to modify forecasts
satellite observations : quality of geostationary and polar orbiting wind data (SATOB), geostationary and polar orbiting brightness temperature data (AMSU-A, HIRS, AMSU-B/MHS and MWHS2, ATMS, AIRS, IASI, CRIS, SSMI/S, GMI, AMSR2, MWRI and GEORAD channels), sea surface wind data (scatterometer), zenithal total delay data (ground GPS stations), radio-occultation data (GPS satellite) and radial speed data (LIDAR) through daily monitoring statistics time series.
Quality of radiances scan position angles retreived from each brightness radiances temperature sensors through monthly monitoring statistics.