Have you ever wondered how much it matters whether a particular data type gets into the model or how much impact on the analysis and forecast a particular data source has? The answer depends on many factors and varies from one case to another, but there are some useful general principles we can consider to help in understanding the impact of data on the model. For a first crack at what these impacts actually look like, Zapotocny et al. (2000) ran a 12-hour EDAS assimilation cycle and 48-hour forecast with all data and separately ran it eliminating each data type. This experimental set-up allows us to see the effect each data source has on the analysis and forecast for the one case that was tested, though it does not allow us to see the combined impact of several data sources. For further details, see their article in the October 2000 issue of Weather and Forecasting.
Some important points about how data affect an analysis and forecast are drawn out here through the answers to three questions utilizing some results from the Zapotocny et al. case study. These lessons apply fairly generally to other cases and other assimilation systems, though some details of the results in Zapotocny et al. may not. The Eta model and 3D-VAR used operationally have been upgraded and are not the same as used in this study. In particular, the 3D-VAR implemented in June 2001 assimilates satellite radiances directly instead of using retrieved satellite soundings, some of the assumed observation error standard deviations are different (affecting how closely the analysis draws to the observations), and the balance constraint linking winds and temperatures is handled in a more sophisticated manner.
You may find it helpful to review data assimilation concepts in the COMET NWP module (link in menu list on left panel), particularly if you have not yet gone through the module.
By Dr. Stephen Jascourt, UCAR/COMET
Reference
Zapotocny, T. H., S. J. Nieman, W. P. Menzel, J. P. Nelson III, J. A. Jung, E. Rogers, D. F. Parrish, G. J. DiMego, M. Baldwin, and T. J. Schmit, 2000: A case study of the sensitivity of the Eta Data Assimilation System. Wea. Forecasting, 15, 603-621.