One goal of the THORPEX program is to improve the design of the global observational network, in order to curtail errors in forecasts in the medium-range and beyond. To achieve this, it is useful to understand the physical basis behind how the assimilation of observations affects global model forecasts, and the factors that compromise their ability to influence forecasts. We explore the influence of 60 independent sets of targeted dropwindsonde observations, released over the north-east Pacific Ocean, on 1-10 day NCEP Global Forecast System forecasts of mid-latitude weather in North America and Europe.

In regard to the dynamical propagation of the effects of the dropwindsonde observations (entitled the "signal"), it is not yet clear whether the dominant component of the signal is advected by the upper-level flow in the storm track, or whether it propagates downstream at the group velocity of Rossby waves, consistent with the dynamics of "downstream baroclinic development". As the signals propagate downstream, they typically spread out and grow slowly over a period of 10 days. In a minority of cases, the signal may be cleanly tracked out to Europe in over a week.

Critically, the useful information in the propagating signal may be degraded due to the contamination by local signals that grow spontaneously and rapidly in certain regions. The initial value of the rapidly-growing component of the signal may be several orders of magnitude smaller than the initial signal at the dropwindsonde locations. The most rapid signal growth occurs in convective areas in the tropics, increasing in scale and then spreading to the mid-latitudes, mostly through advective processes. In the mid-latitudes, the local spontaneous development of signals is most rapid ahead of an upper-level developing trough, typically in the region above the attendant surface low extending across the warm front, coincident with areas of broad ascent and precipitation. An excellent predictor for the location of the development of locally growing signals is regions of CAPE. We conclude that the smallest scales of motion, unresolved by global models, may act to reduce the useful information produced by assimilating observations.