Data Assimilation Problems

What was the cause of the problems with predicting the snowstorm of 6-7 January 2002? It turns out that the models with poor performance had difficulties capturing the initial atmospheric state!

The graphics below show the comparison of the Global Data Assimilation System (GDAS) and the Eta Data Assimilation System (EDAS) to observed radiosondes at 500 hPa at 12 UTC 6 January 2002. The thin black barbs indicate the model initialized 500-hPa winds between the radiosonde observations, while the thick red barbs indicate the raobs. Model initial heights are contoured, while the raob heights are shown with blue text for each reporting station. Note that not all raobs with heights had accepted wind reports.

In the radiosonde observations over the southeastern US, the radiosonde wind at Peachtree City, GA was from 190° at 100 kts., and was from 210° at only 20 kts at Birmingham, AL. Tremendous shear and curvature vorticity is indicated by the radiosondes. One can analyze a short-wave trough axis from these winds running through western TN and along the AL/MS border.

Both 12 UTC analyses, on the other hand, seem to have trouble with both the positioning and the strength of this short-wave trough and the surrounding features. Both:

• Position the trough axis too far east
• Have too little curvature and horizontal wind shear at the base and just downwind from the trough axis, and
• Have too little diffluence downstream of the trough axis.

In spite of not being as low with the 500-hPa heights, however, the Eta initial state seems to have a marginally better depiction of the horizontal shear and curvature in the wind field, and in an upstream short-wave axis from east TX through AR. The subsequent evolution after 12 UTC 6 January in the operational models is not the focus of this case. It appears that microphysics, model resolution, vertical coordinate system, and dynamics may have all played a role in the differing operational model forecasts.

Note that the errors seen here, while not as large, are reminicent of those found in the January 25, 2000 case covered here, where a data assimilation problem resulted in a serious forecast bust. In the current case, the resulting forecast bust was of less magnitude but in the same sense; NWP forecasts indicated a storm track further east than what verified. Significant QPF and snow accumulation errors resulted.