Intercomparison of GNSS-RO Quality Control Checks in NWP. Part II: Superrefraction Tests

Numerical weather prediction (NWP) centers around the world implement different methods from each other in the quality control (QC) of Global Navigation Satellite System radio occultation (GNSS-RO) observations. This study focused on the implementation and evaluation of the superrefraction (SR) QC methods of RO observations used for data assimilation by NWP centers. This study was conducted within the Joint Effort for Data assimilation Integration (JEDI) framework that contains generic QC filters and facilitates easy implementation of new QC methods. This intercomparison of SR QC methods is based on the same framework, i.e., the Met Office forecast model and bending angle forward operator, and a common RO observation dataset. It demonstrates that different SR QCs behave very differently. The methods used by the U.S. Naval Research Laboratory and Météo France remove the most observations, those used by the Met Office and the U.S. National Centers for Environmental Prediction remove the next most, and a method based on differences in impact parameter remove the least. It is noted that some NWP centers implement their QC in a conservative practice aiming at screening out potentially suspicious data. RO observations of sharp refractivity gradients may be rejected even when the observations themselves are of good quality. Comparisons of the vertical refractivity gradients from the observations and the model highlight systematic differences, with sharp refractivity gradients being largely absent from the observations, indicating that QC outcomes depend also on the upstream processing procedures. We hope this study can provide guidance for reconsidering RO QC in NWP practice.