Challenges and progress in assimilation of satellite data in limited-area, rapidly updating modeling systems

Assimilation of satellite data in limited-area, rapidly updating modeling systems poses unique challenges compared to global models systems. Principal among these is the severe data restriction posed by the short data cutoff time (~30 min. for hourly updated systems like the Rapid Refresh). Also, the limited extent of the model domain further reduces spatial extent of satellite data coverage. These two factors impact the bias correction procedures, making them potentially less affective. Within the Rapid Refresh (RAP) hourly updated prediction system, satellite radiance data are assimilated using the Gridpoint Statistical Interpolation (GSI) analysis package. Retrospective RAP experiments have also been conducted, in which AIRS single-field of view (SFOV) retrieved temperature and moisture profiles are assimilated.

For both AIRS radiance and retrieval assimilation, bias correction is a key factor in obtaining forecast improvement. For the AIRS retrievals (obtained from CIMSS, University of Wisconsin), specific bias correction procedures have been developed, based on objective comparison of the retrieved temperature and moisture profiles with nearby (in space and time) radiosonde profiles as well as the WRF background. Application of these bias correction procedures, along with judicious selection of assimilation layers (800-400 mb), assumed observations errors, and horizontal and vertical data filtering have resulted in slight positive impact from assimilation of these data. Particularly noteworthy has been an improvement in wind forecasts (an induced field) from assimilation of the retrieved temperature and moisture profiles.

Companion work has focused on improving the treatment of radiance assimilation within the RAP, both specifically for AIRS radiances and in general for all standard radiance data. For the AIRS radiance data, a Jacabian / adjoint analysis was completed to determine which channels should not be assimilated because of the low (10 mb) RAP model top. Retrospective tests have confirmed that removal of these channels does in fact further improve the already positive impact from assimilation of these data. More recent tests have focused on assimilation of a mixture of radiance data and examining the evolution of the GSI bias correction predictor coefficient fields during a multi-week spin-up period. For some channels, reasonable convergence of the bias predictors is noted, but not for others. Reasons for this are currently being examined. This is a prelude to more extensive systematic tests of the impact of radiance data assimilation for various instruments and channels. Further experiments to evaluation different assimilation strategies are planned and will be reported on at the conference. One potential strategy is to concentrate the radiance assimilation within the RAP partial cycle period, which will allow for a longer data cutoff period, increasing the amount of satellite data included within the RAP assimilation. At the conference, we will report on progress to date in all aspects of this work.