Estimation of Multivariate Observation-Error Statistics for AMSU-A Data

Advanced Microwave Sounding Unit A (AMSU-A) observation-error covariances are objectively estimated by comparing satellite radiances with radiosonde data. Channels 6–8 are examined as being weakly dependent on the surface and on the stratosphere above the radiosonde top level. Significant horizontal, interchannel, temporal, and intersatellite correlations are found. Besides, cross correlations between satellite and forecast (background) errors (largely disregarded in practical data assimilation) proved to be far from zero. The directional isotropy hypothesis is found to be valid for satellite error correlations. Dependencies on the scan position, the season, and the satellite are also checked. Bootstrap simulations demonstrate that the estimated covariances are statistically significant. The estimated correlations are shown to be caused by the satellite errors in question and not by other (nonsatellite) factors.