Shay Gilpin(1,2), Richard Anthes(2), Sergey Sokolovskiy(2), and Therese Rieckh(2)
(1) Significant Opportunities in Atmospheric Research and Science (SOARS), University Corporation for Atmospheric Research, Boulder, CO
(2) COSMIC Program Office, University Corporation for Atmospheric Research, Boulder, CO
Radio occultation (RO) is an active remote sensing method that provides vertical profiles of atmospheric properties. In the retrieval process, RO provides bending angles (BA) of radio wave trajectories traversing the atmosphere between two satellites. The BA profiles (functions of impact parameter) have global coverage, high vertical resolution, little to no instrument bias, and can be directly assimilated into numerical weather prediction (NWP) models. Previous studies have shown assimilation of BA into models has a positive impact on NWP; RO is typically one of the top five observational systems contributing to forecast accuracy. Since BA are not variables analyzed and predicted by NWP models, a forward model must be used to compute BA from model variables (temperature, water vapor, and pressure) to compare with the observed BA in the data assimilation process. The BA forward modeling process requires vertical differentiation of refractivity (a function of temperature, water vapor, and pressure), which is specified on a relatively spare vertical grid. Accurate differentiation requires interpolation of refractivity from the low-resolution model grid to a fine-resolution computational grid. In contrast, the RO observed BA, which are specified on a relatively high-resolution grid, require filtering before comparison to the forward-modeled BA in order to reduce aliasing effects. Thus interpolation and filtering are important steps in the data assimilation process.
In this study, we tested different interpolation schemes (e.g. linear, log-linear, log- spline) from an arbitrary model grid to a high-resolution computational grid to investigate the sensitivity to vertical interpolation and filtering and determine an optimum method that can be applied to any model grid (NCEP GFS, ECMWF, WRF, etc.). The forward-modeled BA are compared to observed BA, which are low-pass filtered with a resolution consistent with the model grid. The root mean square of the difference is used as the criterion of optimality.