NASA's Atmospheric Infrared Sounder (AIRS) has the ability to provide atmospheric temperature and water vapor information at higher resolution and accuracy than previous systems, which may be very beneficial for improving forecasts of high impact weather and cloud and precipitation systems. Accordingly, I am conducting a series of tests to evaluate the impact of assimilating AIRS single field-of-view (SFOV) retrieved temperature and water vapor profiles, which were created by the Cooperative Institute for Meteorological Satellite Studies (CIMSS) at the University of Wisconsin-Madison, as well as raw AIRS radiance data into the NOAA Rapid Refresh (RAP) mesoscale model system. The RAP is a high frequency (1-hour) cycling assimilation and prediction system that replaced the Rapid Update Cycle (RUC) at the National Centers for Environmental Prediction (NCEP) in May 2012.

For the SFOV retrieval assimilation tests, I first assessed the retrieval profiles by comparing them with nearby (in space and time) radiosonde profiles. An initial set of experiments, in which I tested a variety of different data thinning and observation error specification strategies, yielded mostly neutral forecast improvement relative to a control run experiment, in which only conventional observation were assimilated. I then analyzed the observation innovations (O-B), documenting systematic differences between the RAP background fields and the SFOV retrievals, including a systematic dry bias (relative to the background) in the SFOV moisture observations. Based on these results, a simple moisture bias correction scheme was developed and used in a SFOV moisture assimilation experiment, yielding substantially improved results. A more complex vertically and diurnally varying temperature bias correction procedure was then developed, again yielding improved forecast results. Using both bias corrections yielded the best results in RAP retrospective experiments and also produced positive impact in a few selected High Resolution Rapid Refresh (HRRR) runs initialized from RAP parent grids. For one case in particular, the SFOV data reduced an area of anomalously high mid-level moisture in the RAP, leading to a reduction in spurious convection within the 3-km thunderstorm-resolving HRRR model forecast.

AIRS radiance assimilation experiments have focused on channel selection issues associated with the relatively low RAP model top (10 mb), as well as bias correction issues, using the variational radiance bias correction scheme within the Gridpoint Statistical Interpolation (GSI) package. Compared to global models, polar-orbiting satellite radiance bias correction is more difficult for rapidly updating mesoscale models, due to reduced and irregular data coverage. In addition, the relatively low RAP model top (10 mb) compromises the ability to optimally use all AIRS channels. To examine the model top issue, I analyzed temperature and water vapor Jacobians for the various channels. Based on this analysis, channels were then selected for exclusion in the radiance assimilation. Testing of this procedure yielded very good results, with smaller forecast errors (especially aloft) as well as slight improvement for heavy precipitation thresholds for retrospective runs that omitted the identified channels. The bias correction work has focused on evaluating the impact on the forecast skill from the method of initializing the cycled bias correction coefficients and examining the length of time needed to adequately spin-up these coefficients. Preliminary results from a test with a 2 month spin-up indicate a significant forecast improvement compared to no spin-up. Results from all these experiments will be described as well as future work toward operational use of AIRS data within the RAP.