Satellite radiance observation error inflation and spatial thinning: Reconciling idealized theory with near‐operational applications

Abstract This study investigates the joint effects of spatial thinning and observation error standard deviation inflation in satellite radiance assimilation, focusing on the common operational setting in which true observation errors are spatially correlated but are assumed to be uncorrelated. Idealized diagnostics are combined with near‐operational experiments from a global deterministic prediction system to reconcile theory and operational practice, providing a unified, physically based interpretation of thinning–inflation interactions and highlighting the underlying mechanisms governing their relationship. An idealized data assimilation framework, scaled to operationally relevant spatial characteristics, is used to diagnose how optimal inflation depends on thinning distance and on background and observation error correlation length scales. Results reveal a robust asymptotic behavior: when both background and observation error correlation length scales exceed the thinning distance, the optimal inflation approaches the square root of the ratio of the thinning distance. Spectral diagnostics explains this behavior and clarifies why the inflation saturates, providing a physical basis for commonly observed empirical tuning values. Conversely, when the observation error correlation length scale is short relative to the thinning distance, the optimal inflation remains close to unity. These theoretical results are evaluated in a near‐operational context using 15 experimental configurations over a 2.5‐month period, with thinning distances of 150 and 100 km and a range of inflation factors. Reducing thinning without inflation improves the analysis fit to radiances while degrading the fits to several other assimilated observation types and forecast performance, suggesting an excessive influence of the radiances. Increasing the inflation by factors of about 1.4–1.5 mitigates this behavior. Forecast scores demonstrate statistically significant improvements when thinning reduction is paired with appropriate inflation, particularly in the tropics and Southern Hemisphere. Overall, near‐operational results are broadly consistent with the mechanisms identified in the idealized framework, suggesting that these remain relevant in more realistic assimilation systems despite their greater complexity.