A global forecast model "truth run" was conducted at high horizontal resolution. The initial condition was then truncated to a lower resolution (hereafter called the resolved scales), thereby eliminating "unresolved scales." Subsequently, the lower-resolution version of the model was run using the initial condition with only the resolved scales. The effects of the unresolved scales upon the resolved scales can then be determined by comparing forecast output from the resolved scales in the high resolution simulation to the resolved scales in the low resolution simulation.

At the conference the statistical characteristics of the model error contributed by the unresolved scales will be described. Does the model error project significantly on the singular vectors of the system? What are the spatial, temporal, and synoptic characteristics of the model error?

This research is potentially important for ensemble forecasting; typically, ensemble forecasts do not have enough spread amongst the forecasts, and the primary culprit is the unrepresented effects of model errors. Future ensemble forecast systems ought to include realistic methods for incorporating the growing uncertainty due to imperfections in the forecast model itself. This research will provide some understanding of the statistics of one source of model uncertainty.