Satellite Assimilation Activities for the NRL Atmospheric Variational Data Assimilation (NAVDAS) and NAVDAS-AR Nancy Baker 1 , Bill Campbell 1 , Jim Goerss 1 , Rolf Langland 1 , Randy Pauley 4 , Steve Swadley 2 , Tom Rosmond 3 , Liang Xu 1 1 Naval Research Laboratory, Monterey, CA USA. 2 METOC Consulting, Monterey, CA U.S.A 3 SAIC, Monterey, CA U.S.A. 4 Fleet Numerical Meteorology and Oceanography Center, Monterey, CA U.S.A. Introduction The U.S. Navy’s three-dimensional variational analysis system NAVDAS became operational at Fleet Numerical Meteorology and Oceanography Center (FNMOC) on October 1, 2003, paving the way for the direct assimilation of satellite radiances with the U.S. Navy’s global 1 and mesoscale 2 numerical weather prediction models. AMSU-A radiance assimilation, which became operational at FNMOC on June 9, 2004, significantly improved the forecast skill: the two- to five-day forecast skill at 500 hPa increased by 3-10 hours in the Northern Hemisphere (Fig. 1a) and by 12-20 hrs in the Southern Hemisphere (Fig. 1b), compared to ATOVS retrieval control runs. Tropical cyclone track prediction skill also increased at all forecast ranges out to 5 days. Subsequent observation system experiments demonstrated that assimilation of ATOVS retrievals actually degrade NOGAPS tropical cyclone track forecasts, while assimilation of AMSU-A radiances improves the track forecasts (see Fig. 2). The AMSU-A assimilation data selection, quality control and operational bias correction procedures are described in Baker et al. (2005). Although not apparent during the initial multiple-month assimilation tests prior to operational implementation, the AMSU-A bias correction procedure reinforces NOGAPS polar stratospheric biases, ultimately leading to occasional rejection of the Antarctic radiosondes by NAVDAS. The problem occurs because NOGAPS forecast errors are correlated with the bias predictors. The operational bias correction method uses seven predictors, as described in Campbell et al. (2004). They are the 1000-300 hPa tropospheric thickness and 200-50 hPa stratospheric thicknesses, modulated by the square of the sine and cosine of the latitude, skin temperature, NOGAPS total precipitable water and the derivative of the cloud liquid water equation (from Grody et al. 1999). The primary culprits leading to the feedback with the model bias were the skin temperature and the sine and cosine weighting of the thicknesses. The sine and cosine weighting gives too much weight to polar observations. The problems with the skin temperature are more subtle, as the stratospheric biases in the analyses are largely due to the assimilation of higher peaking AMSU-A channels (e.g. channels 8-11) that are not sensitive to the skin temperature. Instead, what 1 Navy Operational Global Atmospheric Prediction System (NOGAPS) 2 COAMPS® is a registered trademark of the Naval Research Laboratory International TOVS Study Conference-XV Proceedings 150