In the lower tropical troposphere, the quality of GPS radio occultation (RO) bending angle and refractivity may vary substantially from one profile to another, due to the variations of the horizontal and vertical structure of water vapor. A standard data quality (QC) control check typically uses the departure of the observation from the first guess, which is based on model forecast, to reject bad observations. Obviously, the effectiveness of such approach depends on the quality of the forecast. The QC procedure is not based on the actual quality of the data, which is often unknown. Also, for the assimilation of observations into a NWP system, appropriate observational errors for a given type of observation must be specified. Such observation errors are often statistically determined. Again, not taken into account the accurate quality of the individual data, which may vary from one profile to another. Given the significant variations of water vapor in the lower tropical troposphere, the observational errors calculated from statistical approach are usually fairly large. This poses the challenge that some good quality RO observations may be under-weighted in data assimilation. A better approach for GPS RO data QC and observational errors specification can be developed, if there is additional information on the quality of individual RO sounding profiles.
In this study, we develop a new approach for QC procedure and observational errors specification based on the local spectral width (LSW) of RO, which is a measure of the uncertainty in RO retrieval due to the non-spherically symmetrical irregularity. It is found that the RO observations with high LSW are substantially low-biased, when compared with independent radiosonde observations and ECMWF global analysis, and should be rejected from assimilation. The RO observations with low LSW, however, are of high quality. Their departure from radiosonde and ECMWF global analysis is substantially less. Observational errors, smaller than the typically statistically determined observational errors, should be assigned to such observations.
Assimilation and forecast experiments of COSMIC RO refractivity data with the Weather Research and Forecasting (WRF) model demonstrate that the LSW-based quality control and LSW-derived dynamic error can improve the quality of moisture and wind analysis, and improve track forecast of tropical cyclones.