Enhanced radiance bias correction in the National Centers for Environmental Prediction's Gridpoint Statistical Interpolation data assimilation system

Radiance bias correction is an important and necessary step in the proper use of satellite observations in a data assimilation system. The original radiance bias-correction scheme used in the Gridpoint Statistical Interpolation (GSI) data assimilation system consists of two components: a variational air-mass dependent component and a scan-angle component. The air-mass component is updated within the GSI, while the scan-angle component is updated outside the GSI. This study examines and enhances several aspects of the radiance bias-correction problem. First, a modified pre-conditioning is applied to the bias-correction coefficients and the analysis variables to speed up convergence of the minimization process. A new procedure for applying the modified pre-conditioning in the GSI is utilized. Second, capabilities for detecting any new/missing/recovering radiance data and initializing the bias correction for new radiance data are implemented. A new scheme is proposed and employed to adjust the background-error variances for the bias-correction coefficients automatically, using an approximation of the analysis-error variances from the previous cycle, and to remove the pre-specified predictor scaling parameters. Finally, the capability to perform bias correction for passive channels within the GSI is developed with a new approach. The two-step bias-correction procedure originally used is replaced with a one-step variational bias-correction scheme within the GSI. Experiment results with the GSI-based hybrid ensemble-variational system show that using the modified pre-conditioning leads to a better convergence rate. Moreover, with the one-step scheme, the anomaly correlation of geopotential height at 500mb is neutral in the Northern Hemisphere but improved in the Southern Hemisphere. The root-mean-square (RMS) error of wind is comparable to that of the two-step scheme and the biases of the global temperature 24 h and 48 h forecasts fitted to the rawinsonde are reduced.