Adaptive observations have been proposed as a means of improving short-to-medium range forecasts of particular weather events by enhancing analyses through "adapting" observing systems to take additional observations in time and space. Adjoint sensitivity fields and singular vectors have been used in defining those regions in which adaptive observations might be taken to improve a forecast. Typically, the response function used to define the adjoint sensitivity has been some measure of cyclone intensity - like a "total" energy norm. While this choice of norm may be desirable for cyclone intensity, it may not be suitable for possibly unrelated aspects of a cyclone like the distribution or intensity of vertical motion and concomitant precipitation, or the distribution of near surface temperature.

An example of such an event occured with the 24-25 January 2000 East Coast cyclone. During that event, a major winter storm deposited as much as 20 inches of snow throughout portions of the eastern United States. This event is noteworthy as operational numerical weather prediction (NWP) model guidance for this cyclone was particularly poor, especially for lead times of 36h and longer. For shorter forecast lead times, while model guidance regarding the surface cyclone position and intensity at 1200 UTC on 25 January 2000 improved, guidance for the precipitation distribution remained poor. With lead times as short as 24 hours, forecasts of the surface cyclone and its associated precipitation were too far to the east. In this example, it is possible that the choice of a response function for cyclone intensity may suggest a different adaptive observing strategy than one formulated based on precipitation distribution.

In this presentation, we calculate the sensitivities of a variety of response functions with respect to initial conditions for the January 2000 case as well as others of interest. The response functions will be related to measures of cyclone intensity, precipitation distribution, and distribution/intensity of forcings for vertical motion. We interpret the differences in these forecast sensitivity fields based on their projections onto the leading singular vectors (defined for the energy norm) of time evolving basic states for the cases studied. We further relate the sensitivity patterns to the underlying synoptic flow regime and will discuss the implications these results have on adaptive observing strategies.