An Examination of the Relative Forecast Error Reduction due to AMVs and Radiances in the Navy Global Environmental Model (NAVGEM)

As part the 11th International Winds Working Group Meeting (February, 2012; Auckland, New Zealand), multiple NWP centers compared the impact of Atmospheric Motion Vectors (AMVs) on their respective numerical weather prediction system analysis and forecasts. The 6-week experiments spanned two seasons (winter, summer). Along with standard forecast error metrics, several NWP centers also produced adjoint-based observation impact (also known as forecast sensitivity to observations, or FSO). These observation impact calculations showed that, for the NRL global atmospheric prediction system, the 24-hr forecast error reduction (computed using a global tropospheric moist error norm) was dominated by AMVs. These results contrasted with those presented by other NWP centers, where radiances dominated the observation impact in their global NWP systems.

These results were subsequently examined in greater detail through collaboration with scientists from NASA GMAO and summarized in a JCSDA Quarterly Newsletter article by Baker et al. (Volume 39, June 2012). The study examined several possible reasons for the larger impact including (1) the greater number of satellite wind data assimilated in the NRL system, (2) the NRL data selection and quality control procedures, especially the use of carefully averaged “super-ob” wind vectors, and (3) the influence of other data types in the assimilation system, including the greater number of satellite radiances assimilated in the GMAO system.

The NRL AMVs (both with and without “super-obs”) were assimilated in the GMAO system, and these results suggested that the greater volume of the NRL AMVs was primarily responsible for their larger impact, although there was evidence that the super-ob processing was beneficial. It was also noted that increasing the number of assimilated AMVs decreased the impact of other leading observing systems, such as satellite radiances, radiosondes and aircraft.

This study re-examines several possible reasons for the relatively larger impact from AMVs and smaller impact from radiances, with a focus on the optimal mix of various observation types, and the role of the observation and background error covariances.